Molecular Vision 2023; 29:217-233 <>
Received 18 August 2022 | Accepted 23 October 2023 | Published 25 October 2023

Protein modeling and in silico analysis to assess pathogenicity of ABCA4 variants in patients with inherited retinal disease

Senem Cevik,1 Nutsuchar Wangtiraumnuay,2 Kristof Van Schelvergem,3 Mai Tsukikawa,4 Jenina Capasso,5 Subhasis B. Biswas,1 Barry Bodt,6 Alex V. Levin,5 Esther Biswas-Fiss1

The first three authors contributed equally to this study.

1Department of Medical and Molecular Sciences, University of Delaware College of Health Sciences, Newark, DE; 2Department of Ophthalmology, Queen Sirikit National Institute of Child Health, Bangkok, Thailand; 3Department of Ophthalmology, Hôpitaux Iris Sud Bruxelles, Bruxelles, Belgium; 4Department of Ophthalmology, Duke University, Durham, NC; 5Departments of Ophthalmology and Pediatrics, Flaum Eye Institute and Golisano Children's Hospital, University of Rochester, Rochester, NY; 6College of Health Sciences Biostatistics Core Facility, University of Delaware, Newark, DE

Correspondence to: Esther E. Biswas-Fiss, University of Delaware College of Health Sciences, Department of Medical and Molecular Sciences, Ammon Pinizzotto Biopharmaceutical Innovation Center 590 Avenue 1743 • Newark, DE 19713; Phone: (302) 831-2912, FAX: (302) 831-4180; email:


Purpose: The retina-specific ABCA transporter, ABCA4, plays an essential role in translocating retinoids required by the visual cycle. ABCA4 genetic variants are known to cause a wide range of inherited retinal disorders, including Stargardt disease and cone-rod dystrophy. More than 1,400 ABCA4 missense variants have been identified; however, more than half of these remain variants of uncertain significance (VUS). The purpose of this study was to employ a predictive strategy to assess the pathogenicity of ABCA4 variants in inherited retinal diseases using protein modeling and computational approaches.

Methods: We studied 13 clinically well-defined patients with ABCA4 retinopathies and identified the presence of 10 missense variants, including one novel variant in the ABCA4 gene, by next-generation sequencing (NGS). All variants were structurally analyzed using AlphaFold2 models and existing experimental structures of human ABCA4 protein. The results of these analyses were compared with patient clinical presentations to test the effectiveness of the methods employed in predicting variant pathogenicity.

Results: We conducted a phenotype–genotype comparison of 13 genetically and phenotypically well-defined retinal disease patients. The in silico protein structure analyses we employed successfully detected the deleterious effect of missense variants found in this affected patient cohort. Our study provides American College of Medical Genetics and Genomics (ACMG)–defined supporting evidence of the pathogenicity of nine missense ABCA4 variants, aligning with the observed clinical phenotypes in this cohort.

Conclusions: In this report, we describe a systematic approach to predicting the pathogenicity of ABCA4 variants by means of three-dimensional (3D) protein modeling and in silico structure analysis. Our results demonstrate concordance between disease severity and structural changes in protein models induced by genetic variations. Furthermore, the present study suggests that in silico protein structure analysis can be used as a predictor of pathogenicity and may facilitate the assessment of genetic VUS.


Found primarily in the outer segments of vertebrate rod and cone photoreceptor cells, ABCA4 plays an essential role in phototransduction. A large number of variants in the ABCA4 gene are known to be associated with a wide range of autosomal recessive inherited retinal disorders (IRDs), including Stargardt disease (STGD1, OMIM 248200), cone-rod dystrophy (CORD3, OMIM 604116), fundus flavimaculatus (FFM), and retinitis pigmentosa (RP, OMIM 268000). To date, more than 3,000 variants in the ABCA4 gene have been reported (ClinVar; accessed April 23, 2023), with ABCA4 representing the most frequently mutated gene associated with human retinal dystrophy [1]. ABCA4 variants include missense, nonsense, frameshift, splicing, and structural mutations, with missense variants constituting 46% [2].

The ABCA4 gene consists of 50 exons coding for a 2,273-amino-acid polypeptide chain. The ABCA4 protein consists of two tandem-arranged halves, each consisting of six membrane-spanning α helices, a large and evolutionary-conserved extracytoplasmic domain (ECD), a cytoplasmatic nucleotide-binding domain (NBD), and a regulatory domain (RD; Figure 1). Adenosine triphosphate (ATP) binding and hydrolysis occur in the two symmetric NBDs (NBD1 and NBD2), while ECD1 and ECD2 are suggested to play a role in retinal substrate binding [3,4].

Genetic variation databases such as ClinVar [2], Online Mendelian Inheritance in Man (OMIM) [5], the Human Gene Mutation Database (HGMD) [6], and the Leiden Open Variation Database (LOVD) [7] are widely used in clinical settings to infer a variant’s phenotypic consequences. They help clinicians to predict disease prognosis and, possibly, to guide patients to suitable clinical trials when available. The jointly accepted standard terminology for interpreting the gene variants in Mendelian diseases is the five-tier classification: benign, likely benign, uncertain, likely pathogenic, and pathogenic [8]. At least half of the missense ABCA4 variations have been designated as variants of uncertain significance (VUSs), and 12% of the remainder have conflicting interpretations (CIs; ClinVar; accessed April 23, 2023). Thus, elucidating ABCA4 VUSs is a critical step in the progress of precision medicine in retinal diseases.

A remaining challenge in ABCA4-opathies is pinpointing the causative variants and assessing the combinatorial effects of biallelic variants [9-12]. Although some recurring variants of ABCA4 have been characterized in terms of their effects on protein function and, thus, their predisposition to pathogenicity, experimental studies of the extensive and growing numbers of nonsynonymous ABCA4 variants may be infeasible. Only a fraction of ABCA4 variants have been functionally studied [13-20]. Genotype–phenotype correlation for ABCA4-related retinal degenerative disease has proved challenging due to the heterogeneity in both the phenotype and genotype and in the large number of reported sequence variations. Relatively few studies have reported on specific gene variants that are correlated with specific phenotypes [9,17,21-24].

Thus, a robust platform is needed to predict the molecular consequences of ABCA4 variants. In silico studies may provide a suitable predictive model for the role of ABCA4 variants in retinal diseases. Although several pathogenicity prediction tools are available, most of them conduct their assessments based primarily on sequence features and/or evolutionary conservation and do not explain the possible functional effects at the protein level. The results of in silico studies can be enhanced using a structure-based pathogenicity analysis of ABCA4 variants, which may lead to an improved predictive pathogenicity model.

A comprehensive study testing the reliability of three-dimensional (3D) protein structures to assess the pathogenic impact of missense variants concluded that structural damage could be accurately predicted using computational protein models [25]. Several studies have successfully used computational protein modeling to assess variations’ pathogenic impact on various conditions, including developmental, neurologic, and metabolic disorders [26-33]. Our recent in silico study found concordance between predicted protein structural changes and known pathogenicity classification in a set of ABCA4 clinical variants [34].

In this study, we further assessed the ability of computational protein structure analyses to predict the pathogenic impact of ABCA4 variants in a phenotypically well-defined cohort of inherited retinal disease patients. Using AlphaFold2 (AF2) protein modeling [35] and available cryogenic electron microscopy (cryo-EM) structures of the human ABCA4 protein, we identified an alignment between severe clinical phenotypes and alterations in protein structure caused by ABCA4 mutations.



This retrospective study was conducted in accordance with Institutional Review Board approval (IRB#14–403E) of the Wills Eye Hospital and Thomas Jefferson University. A retrospective chart review was conducted for those patients with retinal disease–harboring mutations in the ABCA4 gene. All 13 patients were seen in the Ocular Genetics Clinic at Wills Eye Hospital from August 2010 to August 2015 by an ocular genetics specialist (AVL). All patients who were found to have only a heterozygous ABCA4 mutation or who had another suspected gene that might have modified the phenotype were excluded from this study. The medical records were reviewed, and the following data were collected: age of onset, age first seen, best-corrected visual acuity, and results of slit lamp examination and dilated fundus examination. Ophthalmic diagnostic investigations were reviewed, including fundus autofluorescence (AF), spectral-domain optical coherence tomography (SD-OCT), Goldmann visual field (GVF), and, if done, intravenous fluorescein angiography (IVFA). Full-field electroretinogram (ffERG) and multifocal electroretinogram (mfERG) were performed in accordance with the International Society of Clinical Electrophysiology and Vision standards [36]. Testing was customized for each patient as clinically indicated, based on age, visual acuity, and other factors. No clinical testing was done specifically for this study.

Genetic testing

The ABCA4 gene was screened for variants using a range of molecular diagnostic methods, in isolation or in parallel with other genes, as determined from the phenotype and considering the patients’ insurance coverage for testing (Appendix 1). When clinically indicated testing could not be performed—for example, due to a lack of insurance coverage—research testing was offered with the confirmation of detected variants performed in a Clinical Laboratory Improvement Amendments (CLIA)–certified laboratory.

Protein modeling and pathogenicity prediction

To predict the structural and functional consequences of missense ABCA4 variants, protein structure analysis was performed on AF2-predicted protein models and the existing experimental structures of the human ABCA4 protein [4,37,38]. Full-length (FL) and domain-specific protein modeling was performed using AF2 [35] for wild-type (WT) proteins and each variant protein separately, as described previously [34]. We used AF2 through the ColabFold notebook in the Google Colaboratory [39], which is reported to have a level of accuracy that is almost identical to that of AlphaFold v2.3.1 locally for monomeric structures [36]. The default parameters on the ColabFold notebooks (AlphaFold2.ipynb and AlphaFold2_advanced.ipynb) were used, and the highest confidence models (rank 1) were selected among the generated models. The predicted WT and variant structures were refined for energy minimization using the Amber-Relax option in ColabFold, and the “repair object” feature in the FoldX plugin in YASARA View (free version: 22.5.22) [40,41]. The refinement process was repeated several times until no movement of the side chains was observed in each model. PyMOL2 (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC) was used for model visualization and structure analysis.

The structure analysis of variants was performed using two approaches. First, we compared the WT and variant AF2 models to assess 3D conformational changes, to calculate the root-mean-square deviation (RMSD) alignment scores, and to identify potential impairments in secondary structure elements. Second, we evaluated the substitutions on the ABCA4 cryo-EM structures, considering various aspects, including stereochemical properties, binding, and molecular interactions (i.e., polar content, weak interactions, surface charge, and accessibility, cavities, and pockets). We measured the all-atom RMSD between the AF2-generated FL WT model and each variant, employing the “align” command without outlier rejection (cycles = 0) in PyMol2. We analyzed clashing interactions in available experimental structures and the refined AF models, as steric clashes resulting from the unnatural overlapping of newly positioned side chains with other residues can lead to conformational changes and structural destabilization due to a mutation.

To predict stability changes in protein variants, we used the FoldX plugin for the YASARA software. The ΔΔG free-energy change, calculated as the difference between the folded and unfolded free-energy states (ΔGmutant – ΔGwild-type), was determined using the “mutagenesis” feature of the software [40,41]. We reported the ΔΔG values obtained from all available ABCA4 experimental structures.

Furthermore, we evaluated the suitability of the AF2-predicted FL ABCA4 model for this stability analysis, by comparing the ΔΔG values per variant calculated using the AF2 model with those based on the experimentally determined structures. To facilitate this comparison, we conducted a Spearman’s rank correlation using the JMP statistical software package (JMP®, Version 17, SAS Institute, Inc., Cary, NC, 1989–2021). It is important to note that in silico ΔΔG values are approximate indicators of potential stability changes rather than absolute, experimentally determined values.

For the informatics-based pathogenicity prediction of missense variants, we used Combined Annotation Dependent Depletion (CADD) [42], Polyphen-2 [43], REVEL [44], and ConSurf [45] for the evolutionary conservation of the amino acids. We also used Human Splicing Finder (HSF-Pro) Version 4 to predict the possible splicing effects of all variants in the study [46].


Patient characteristics

We enrolled 13 patients diagnosed with retinal dystrophy caused by variants in the ABCA4 gene. Segregation analysis was performed for all patients to ascertain that the patients were biallelic for variants in the ABCA4 gene. To our knowledge, the patients did not harbor any other pathogenic variants in other confounding retinopathy genes. Genotype–phenotype comparisons were conducted for the ABCA4 variants occurring in these patients.

The clinical characteristics of the patient cohort are summarized in Table 1. The age first seen ranged from 6 to 65 years old, with a median age of 14 years old. The age of onset ranged from 6 to 15 years old, with a median age of 12 years. The best-corrected visual acuity varied from 20/30 to light perception. Two patients (A03M, A08M) had a visual acuity difference between the fellow eyes of more than two Snellen lines. In patients A02M and A05F, eccentric fixation was observed. The extraocular movements and applanation tonometry were normal in all patients. Nuclear sclerosis cataract was found in two patients (A02M and A04F).

OCT showed some degree of photoreceptor loss in 12 patients (92.3%), while 1 patient had a normal OCT. Seven patients were shown to have subretinal deposits with geographic atrophy, and five patients had only subretinal deposits. The foveal thickness in the right eye varied from 69 to 226 μm. Six patients had a normal GVF, three had a constricted visual field, and one had a central scotoma. For three patients, visual field testing was not performed. An ffERG was done in 12 patients, and the results were abnormal for 11 (91.66%). Seven patients had an abnormal response only in the photopic and four patients in both photopic and scotopic ERG. Following an isoelectric ERG, one patient had both abnormal scotopic and photopic responses. For all 12 patients who had an mfERG, abnormal results were produced. IVFA was done in 8 patients, and a silent choroid was found in 5 (62.5%).

There were 3 homozygous ABCA4 mutation patients and 10 compound heterozygous ABCA4 mutation patients. We identified 19 different ABCA4 variants among the 13 patients: 3 frameshift variants, 1 nonsense variant, 5 splicing variants, and 10 missense variants (Table 2). We also identified one VUS, four CIs, and one novel ABCA4 variant in this study, as reported in the ClinVar database (Table 3 and Table 4).

Protein modeling and structure analysis

To compare the deleterious impacts of the missense variants with the patients’ disease severities, we used AF2 protein modeling and all existing human ABCA4 cryo-EM structures (Protein Data Bank identifiers [PDB IDs]: 7lkp, 7lkz, 7e7i, 7e7q, 7e7o, 7e7q, 7m1p, 7m1q). ABCA4 WT and variant models were obtained in high confidence (pLDDT > 80%) and with high structural similarity with the experimental structures (Appendix 2, Appendix 3 and Appendix 4).

Variants inside the open reading frame (ORF) can have various impacts at the protein level. Some variants have clearly devastating consequences, such as those that lead to the misfolding and mislocalization of the ABCA4 or truncating mutations, which result in the complete loss of function (LoF). However, LoF can also result from local effects on the protein structure and stability that are more subtle and less obvious. Computational analysis aids in the evaluation of these types of effects. These types of local damage may lead to a spectrum of function loss, including affected intramolecular and ligand interactions. To interpret these types of functional damage, we used in silico protein analysis. For all missense variants, we examined the conformational changes, secondary structure elements, relative solvent accessibility, RMSD alignment scores, polar content alterations, clashing interactions, surface properties, and Gibbs ∆∆G free-energy changes. In silico analysis found a range of structural alterations in 10 missense ABCA4 variants carried by affected individuals (Table 3, Figure 2).

Of the 10 missense variants analyzed, 9 resulted in substantial structural damage. These variants caused clashing interactions with neighboring residues (p.C54Y, p.L541P, p.P1380L, p.A1598D, and p.G1961E), disrupted secondary structure elements (p.L541P and p.T1253M), replaced a hydrophobic buried residue with a hydrophilic one (p.A1598D and p.G1961E), or affected molecular interactions (p.R212C, p.R1098C, p.P1380L, and p.E2233V). The remaining variant, p.A1038V, led to a local conformational change in the backbone and reduced stability (Figure 2, Table 3). The stability analysis suggested that 6 of the missense variants (p.C54Y, p.L541P, p.A1038V, p.P1380L, p.A1598D, and p.G1961E) were particularly destabilizing with ∆∆G > 5 (Table 3 , Appendix 4).

We examined how the AF2 model determined ΔΔG values compared to those calculated using the ABCA4 cryo-EM structures. The ∆∆G profiles over 10 missense variants (and 2 complex alleles) were formed to compare these measures. Seven values were missing because the Arg-212 residue was unavailable in the 7lkp, 7m1p, and 7lkz structures, and the Gly-1961 and Glu-2233 residues were absent in the 7mlq structure. Appendix 5 shows agreement in the relative ∆∆G values for each missense variant among the experimental structures and the AF2 model. We formalized the similarity seen with a pairwise correlation analysis using Spearman’s rank correlation. Unadjusted p values showed 26 of 28 possible paired correlations to be significant (maximum p value = 0.0114); the ΔΔGs calculated based on AF2 specifically correlated with those done on the experimentally determined structures at approximately 0.89 or above for all except 7lkz (⍴ = 0.75; Appendix 6). Based on these findings, ΔΔG predictions from the AF2 model can be considered an alternative to those obtained from experimentally determined structures.

In silico protein structure analysis and phenotypic association

The missense variants were distributed across the ABCA4 protein sequence: there were 3 in ECD1, 3 in NBD1, 1 in ECD2, 1 in transmembrane domain-2 (TMD2), and 2 in NBD2 (Figure 1, Table 3). The variants identified in our cohort are described in terms of the clinical phenotype and assessment of the impact of the variant on protein structure.

The p.C54Y variant was identified in severely affected siblings, A04F and A05F, with ages of onset of 14 and 13 years, respectively. The two patients had the same biallelic combination, with p.E2233V on the opposite allele (in trans). The visual acuity of these patients was poor, ranging from 20/200 to 20/400. The predicted protein model and experimental structures showed disulfide bond breakage and clashing interactions (Figure 2A), along with a poor alignment score (RMSD = 1.26) and destabilization as reflected by the ∆∆G (+ 41.68 kcal/mol), supporting the deleterious effect of this variant (Table 3). The informatics tools also predicted the variant to be pathogenic (Table 3, Supplementary Table S2). The observed severity of this variant, based on the patient’s phenotype (Table 1), aligned well with the protein structural change.

Variant p.R212C was identified in a 6-year-old patient (A09M) with a rapidly progressing disease. The patient’s visual acuity at the time of the study ranged from 20/80 to 20/200, with constricted visual fields. OCT showed subretinal deposits and geographic atrophy (Table 1). The p.R212C variant was found in trans with the c.768G>T:p(V256=) synonymous variant that was predicted to be pathogenic due to its likely effect on splicing (HSF-Pro; Table 4) [46]. Given that one allele was predicted to be completely compromised (due to coding for a severely truncated polypeptide), the pathogenicity level of the p.R212C allele was expected to be directly related to the patient’s disease severity. The p.R212C variant was predicted to be pathogenic by the informatics tools (Supplementary Table S2), and structure analysis revealed a missing H-bond with Gln-190 due to the substitution (Figure 2B, Table 3). However, p.R212C is in the part of the ECD1 domain where the experimental structures have not been well resolved. This residue might be important for the structural flexibility of the domain.

Complex allele p.L541P/p.A1038V was detected in 3 patients (A06F, A07M, and A08M) in our study. The siblings A07M and A08M had the same biallelic combination with the c.5714+5G>A splicing variant on the trans allele, while the third patient had a p.Q1412* nonsense mutation in trans (Table 2). All three patients exhibited fundus flecks, subretinal deposits, and pigmentary clumps. The age of onset, visual acuity, and other findings were comparable in the siblings (Table 1). Protein structure analysis revealed that p.L541P had a more deleterious impact than did p.A1038V by causing significant impairment to the protein structure. This substitution is located within an α-helical region and is predicted to disrupt the α-helix structure and cause clashing interactions, leading to conformational changes (RMSD = 1.27; Figure 2C, Table 3). Although we did not observe a noticeable structural difference between the WT and p.A1038V variant, it showed an overall poor alignment score (RMSD = 1.25) and ∆∆G (+ 6.56 kcal/mol), indicating that it can be a destabilizing mutation (Figure 2D). The informatics-based tools suggested that the p.L541P variant is pathogenic while p.A1038V is benign (Table S2). For the complex allele L541P/A1038V, the computational analysis of the variant on the ABCA4 protein structure aligned well with patient phenotype.

Variant p.R1098C was found in patient A13M, who had an age of onset of 11. It was in trans with a frameshift mutation (Ala222Glnfs*19), which leads to a complete loss of function. The Arg1098 residue, localized inside the NBD1, was found to have an essential role in the NBD1–RD2 interdomain interaction by forming a salt bridge, which is disrupted by the Cys substitution (Figure 2E). Prediction tools classified p.R1098C as pathogenic (Table 3, Supplementary Table S2). For this variant, the computational analysis of the effects on the ABCA4 protein structure coincided with the disease phenotype.

Complex allele p.T1253M/p.G1961E was identified in patient A03M, with the c.5714+5G>A splicing variant in trans, which is predicted to affect the splicing pattern according to HSF-Pro [46]. The patient presented with a “beaten bronze” appearance in fundus imaging, with lipofuscin deposits centrally and a mild surrounding ring of hyperpigmentation (Table 1). Remarkably, 10 years after the diagnosis, the patient’s vision remained relatively preserved compared to the other patients in our cohort. In silico analysis predicted that both missense variants individually cause protein structural changes. The p.T1253M variant, located within RD1, led to a major structural alteration with a degenerated secondary structure element (Figure 2F). While, the p.G1961E variant replaced a hydrophobic buried residue with a hydrophilic amino acid, which was predicted to destabilize the structure, as indicated by the ∆∆G value (Table 3) and as shown in Figure 2I. The prediction tools unanimously supported the pathogenicity of both variants (Supplementary Table S2).

The p.P1380L variant was identified in a 10-year-old homozygous patient (A010M) with a highly progressive ABCA4 retinopathy. The patient’s severe phenotypic findings (Table 1) are consistent with protein modeling that indicates the Pro → Leu substitution within the transmembrane domain has significant structural consequences. The substitution causes steric clashes with neighboring residues in all possible positions, leading to altered protein conformation (RMSD = 1.26). Additionally, the variant protein model superimposed with the substrate-bound cryo-EM structure suggests that this proline residue likely plays a role in a kinked transmembrane helix, with implications of a functional role in substrate translocation (Figure 2G).

The p.A1598D variant, located in the ECD2 domain, was identified in patient A01F, who presented with macular flecks and alterations in the retinal pigment epithelium (RPE) subretinal deposits, with a visual acuity of 20/200 in both eyes. The patient was homozygous for the p.A1598D variant, with no other ABCA4 variants detected. In silico analysis revealed that by introducing a hydrophilic residue to a buried location, the p.A1598D variant resulted in destabilization (∆∆G = + 13.56 kcal/mol) and caused steric clashes (Figure 2H). However, the conventional pathogenicity assessment of this variant resulted in inconclusive predictions. PolyPhen-2 and REVEL both classified it as possibly damaging, with scores of 0.57 and 0.685, respectively. In contrast, the CADD score of 16.3 was below the threshold of 20, indicating a lower likelihood of pathogenicity. Ala-1598 is not evolutionarily conserved [45], but the variant has a low allele frequency (2.63 × 10−5; Supplementary Table S2) [47]. Overall, the patient’s severe phenotype was consistent with the substantial changes observed in the protein structure predicted by structure analysis.

Variant p.E2233V was identified in siblings A04F and A05F, who had comparable age of onset and visual acuity (Table 1). A04F presented with large retinal pigmentary clumps involving the macula and mild periphery, choroidal attenuation, but no fleck lipofuscin deposits. A05F showed geographic atrophy in the macular area with mid-periphery pigmentary clumping, with several subretinal flecks in the macula. A grayish elevation in the superior macula could represent either a choroidal nevus or subretinal neovascular membrane. Both patients had relatively healthy optic nerves, with minimal pallor as well as mild retinal vascular attenuation (Table 1). This novel variant is located in RD2. Computational analysis revealed a loss of an H-bond with a residue in the NBD1 domain, significantly impairing the protein structure (RMSD = 1.35; Figure 2J). This variant was also predicted to affect splicing by altering the auxiliary sequences’ ESE/ESS motifs ratio (−10; HSF-Pro [46]). Although the true consequence of this variant is not yet fully understood, we saw an agreement between the computational analysis of the missense effects on the ABCA4 protein structure and the patient phenotype.

Our cohort had three patients who did not have a missense ABCA4 variation. Patient A02M carried the c.5018+2T>C variant in a homozygous state. This intronic variant is predicted to affect the splicing pattern by causing a broken WT splice site (HSF-Pro [46],), resulting in translated product lacking the last quarter of the protein. Patient A11F was compound heterozygous with the c.859–13T>C and c.4539+2028C>T intronic variants, present in the ClinVar database as a CI and VUS, respectively. This variant was also predicted to affect splicing by altering the ratio of Exonic Splicing Enhancer (ESE) to Exonic Splicing Silencer (ESS) motifs, with a score of −10 as predicted by the HSF-Pro [46]. The c.4539+2028C>T deep intronic variant has previously been shown to cause 345-nt pseudoexon insertion with a likely effect on the protein level (p.R1514fs) [48]. Unless there are other undetermined variations in the patient, the phenotype suggests that these intronic variants are pathogenic. Patient A12F carried two different frameshift variants—p.I284Vfs*33 and p.V2062Tfs*51—both with high levels of pathogenicity.


The ABCA4 gene, with more than 3,000 reported variants, is known to be associated with a wide range of autosomal recessive inherited retinal disorders. Unfortunately, many of these variants remain uncharacterized in terms of their disease propensity, making it difficult to determine disease prognosis when detected in patients. With the recent advancement of computational tools such as pathogenicity prediction software and protein modeling platforms, it is now possible to carry out detailed in silico investigations of specific variants and to correlate them with available clinical findings, even in the absence of an in vitro functional analysis. In this study, we investigated the use of the in silico analysis of the ABCA4 protein as a tool to augment the prediction of disease severity and the clinical significance of sequence variants.

To evaluate this paradigm, we performed a patient-based study of individuals with confirmed ABCA4 retinopathy. Subsequently, we performed a computational protein structure analysis to see how well the predicted structural changes aligned with the observed pathogenicity of the missense variants based on the patient phenotypes. The phenotypic outcome of ABCA4 retinopathies depends on the collective effect of the specific variants in the patient [12,49,50]. Therefore, we included all types of variants with their predicted effects when discussing the relation between a patient’s genotype and phenotype. Furthermore, we applied a combinatorial approach to increase accuracy in predicting pathogenicity and explaining the pathogenicity of disease-associated variants, as no single type of analysis is sufficient to explain the deleterious effect of a genetic variant at the protein level (Figure 3).

We used AF2, a highly reliable way to predict protein structures [35,51,52], along with recently resolved experimental structures of ABCA4 to analyze the structural impacts of the variants. In addition to physically observable conformational changes, we evaluated bonding, interaction, surface characteristics, and protein stability in the 3D protein models to gain insight into the effects of variants on the ABCA4 protein. We used two numerical measures: the RMSD calculation to compare the conformational alignment scores between the WT and mutant predicted protein structures, and the ∆∆G calculation to compare thermodynamic stability changes upon amino acid substitution. We observed the consistency between the in silico ΔΔG values determined by the AF2 model and the ABCA4 cryo-EM structures (Supplementary Figure S3, Supplementary Table S3). Our findings are in line with a recent study by Akdel et al. [52], supporting the notion that AF2 models can serve as a reliable alternative for predicting the destabilization effects of variants based on protein structure, which can be particularly helpful for proteins with no available experimental structure.

In the present study, we used manual approaches to detect structural alterations; however, with the help of this initial study and with larger data sets, more sophisticated, automated, or machine learning–based strategies such as support vector machines (SVMs) can be used. It should be noted that although the variants alter the protein structure, it does not follow that they will necessarily alter its function [53]. We also acknowledge that the structural changes we mention here are insufficient to explain all the in vivo effects and all aspects of the mutation consequences, including mRNA-level and post-transcriptional alterations. Functionality and localization analyses can be performed to support the findings of this study.

According to American College of Medical Genetics and Genomics / Association for Molecular Pathology (ACMG/AMP) guidelines [8], being detected in the trans to another pathogenic variant provides at least a moderate level of evidence of pathogenicity (PM3) for genetic variants in recessive disorders. Six missense variants in our cohort (c.161G>A:p(C54Y), c.634C>T:p(R212C), c.3292C>T:p(R1098C), c.4139C>T:p(P1380L), c.4793C>A:p(A1598D), and c.6698A>T:p(E2233V)) were known to be alone in the corresponding allele (i.e., no other variants in cis; Figure 4), and all of them were structurally damaging (Figure 2). Our findings from the protein structure analysis support the expected pathogenicity of these six missense variants identified in trans to a pathogenic allele in affected patients of a recessive disorder.

In addition to being identified in the affected individuals who presented with severe phenotypes, six of the missense variants in our study (p.R212C, p.L541P, p.A1038V, p.P1380L, p.A1598D, and p.G1961E) have previously been characterized in functional studies [17,19,20,23,54] and presented with impairments in protein expression, localization, or enzymatic function (PS3; Table 3, Figure 4); this provided an excellent opportunity to assess the employed in silico approach. We found severe structural defects and destabilizing effects in relation to the variants p.L541P, p.P1380L, p.A1598D, and p.G1961E (Figure 2, Table 3), consistent with previous functional studies. Specifically, p.L541P and p.P1380L exhibited mislocalization and a significant reduction in basal and retinal stimulated ATPase activities [17,19]. The variants p.A1598D and p.G1961E, in contrast, showed almost a complete loss of retinal-stimulated ATPase activity [17,20]. These findings suggest that in silico protein structure analysis can predict the functional effects of ABCA4 missense variants.

We had one novel variant, c.6698A>T:p(E2233V), in our cohort. We used the ACMG and AMP guidelines, which provide a series of standard criteria for classifying the potential pathogenicity of variants [8]. Our study provided different degrees of pathogenicity evidence for nine missense ABCA4 variants, including this novel variant (Figure 4). The p.E2233V variant was found in affected siblings in our cohort (supporting evidence of pathogenicity: PP1 and PP4, ACMG/AMP Guidelines). It was alone on the allele and in trans to another pathogenic variant (PM3). The standard informatics tools predicted that its effect is probably damaging (Supplementary Table S2), and protein modeling demonstrated substantial impairment in the structure (PP3). We determined the p.E2233V variant as likely pathogenic based on the supporting evidence, and we deposited it into the ClinVar database (Accession: SCV002526665.1).

Identifying disease-causing variants in inherited diseases is undoubtedly crucial in clinical management, including future clinical trials. In our study, three missense ABCA4 variants had conflicting interpretations in ClinVar. While p.L541P was one of them, it can be considered pathogenic/likely pathogenic, as the conflict arose from only 1 submission with uncertain significance (ClinVar submissions: 13 pathogenic, 2 likely pathogenic, and 1 uncertain significance). The other variants with conflicting interpretations were p.T1253M and p.G1961E, which were found together as a complex allele in patient A03M in our cohort, precluding allelic assessment. However, in silico analysis predicted both variants to be structurally defective, and conventional prediction tools suggested that both were likely pathogenic (PP3). Furthermore, patient A03M carried this complex allele in trans to the pathogenic c.5714+5G>A splicing variant, which was also found in two other patients in our cohort in trans with the p.L541P/p.A1038V complex allele (A07M and A08M), who presented with poorer outcomes (Table 1). These findings suggest that the p.L541P/p.A1038V complex allele may have a worse clinical outcome than does the p.T1253M/p.G1961E complex allele.

Homozygous patients offer a unique opportunity to investigate the genotype–phenotype relationship in autosomal recessive diseases. In our study, we had three homozygous patients. The p.P1380L biallelic variant was found with early onset (i.e., 10 years old) disease presenting with the rapid deterioration of sight. The patient showed a “beaten bronze” appearance on examination, with diffuse flecks and pigmentary clumps. OCT showed subretinal deposits and geographic atrophy. In silico analysis found that the p.P1380L caused severe structural alterations, including conformational change, reduced stability, and likely intermolecular interactions (Table 3, Figure 2G). Biallelic p.A1598D led to a mid-adolescence onset but a progressive disease phenotype with 20/200 visual acuity in both eyes, presenting with flecks and subretinal deposits in the macular area. Similarly, this variant led to decreased protein stability, as indicated by the structure analysis and the ∆∆G value (Table 3, Figure 2H). The last biallelic variant was the intronic c.5018+2T>C variant, found in a patient with onset at 15 years old. Over a disease duration of 50 years, the patient lost his vision to the degree of light perception (Table 1). This variant was predicted to affect splicing by the HSF-Pro web server (Table 4) [46].

In this study, we illustrated the predicted structural consequences of genetic variations found in patients with ABCA4-retinopathies. It is promising to see that the protein modeling and in silico analysis used in this study successfully identified structural defects in variants found in affected individuals, thus allowing for the identification of variants that were expected to be pathogenic. In our recent study, we also demonstrated the accuracy of this in silico approach by assessing known pathogenic and known benign ABCA4 missense variants retrieved from clinical databases [34]. However, it is important to test this in silico approach on a large number of variants found in patients with diverse clinical presentations to observe whether the predicted structural changes are proportional to disease severity. As a transporter protein, ABCA4’s function is closely related to its dynamic properties [55,56], and some sequence variants can impact the dynamic properties of the protein. In this study, we did not assess these potential molecular dynamics–related impacts. Also, we did not conduct deep intronic or ABCA4 promoter sequencing, so the effects of these potential variants, if present, in cis with the known variants as well as the effect of variants in other genes remains unknown.

Our study shows that the types of structural damage caused by the ABCA4 variants found in inherited retinal disease patients are consistent with their affected phenotypes, the assessments of genetic variation databases, previous functional studies, and, for the most part, conventional in silico pathogenicity prediction programs. Our computational protein structure method of analysis helps predict the functional roles of individual amino acids in the ABCA4 protein and provides possible explanations for the pathogenesis of missense mutations. The promising results of this small-scale study suggest that this methodology can be extended to larger and more diverse cohorts. Additional studies with biochemical and molecular focuses can be conducted to better understand all aspects of the variants studied here. We conclude from the results of this study that computational protein structure analysis may be a helpful adjunct in predicting the pathogenicity of ABCA4 VUS and in disease prognosticating.


In silico methods are essential elements of biomedical research. Computational protein structure analysis is a practical tool used to understand protein structure and function, as well as to infer the pathogenic impacts of sequence variations. This study illustrated the likely structural consequences of 10 missense variants. Our study demonstrated that a disease severity that is closely defined by the age of onset and by visual acuity is well associated with in silico–predicted protein structural changes caused by missense ABCA4 variants. Determining disease-causing variants is a crucial part of clinical management, and we believe that our research will contribute to elucidating the molecular pathology of variants associated with ABCA4 retinal diseases.

Appendix 1. Supplementary Table S1. Genetic testing for patients in this study.

Appendix 2. Supplementary Figure S1.

Appendix 3. Supplementary Figure S2.

Appendix 4. Supplementary Table S2.

Appendix 5. Supplementary Table S3.

Appendix 6. Supplementary Figure S3.


Financial support: Supported in part by the Foerderer Fund (A.L.), the Robison D. Harley, MD Endowed Chair in Pediatric Ophthalmology and Ocular Genetics (A.L.), Adeline Lutz - Steven S.T. Ching, M.D. Distinguished Professorship in Ophthalmology (A.L.), an unrestricted grant from the Flaum Eye Institute by Research to Prevent Blindness, and grants from the National Institutes of Health (NEI, R15EY013113; E.B.-F.), and the Foundation Fighting Blindness (BR-GE-0623-0860-UDEL E.B.-F.). We gratefully acknowledge the fellowship grant to S.C. by the Republic of Turkiye Ministry of National Education. The authors thank Dr. Shawn Polson for his valuable discussions related to bioinformatics and Jocelyn Korth for her contributions. Ethics approval: This retrospective study was conducted in accordance with Institutional Review Board approval (IRB#14–403E) of the Wills Eye Hospital and Thomas Jefferson University.


  1. Stone EM, Andorf JL, Whitmore SS, DeLuca AP, Giacalone JC, Streb LM, Braun TA, Mullins RF, Scheetz TE, Sheffield VC, Tucker BA. Clinically Focused Molecular Investigation of 1000 Consecutive Families with Inherited Retinal Disease. Ophthalmology. 2017; 124:1314-31. [PMID: 28559085]
  2. Landrum MJ, Lee JM, Benson M, Brown GR, Chao C, Chitipiralla S, Gu B, Hart J, Hoffman D, Jang W, Karapetyan K, Katz K, Liu C, Maddipatla Z, Malheiro A, McDaniel K, Ovetsky M, Riley G, Zhou G, Holmes JB, Kattman BL, Maglott DR. ClinVar: improving access to variant interpretations and supporting evidence. Nucleic Acids Res. 2018; 46D1:D1062-7. [PMID: 29165669]
  3. Biswas-Fiss EE, Kurpad DS, Joshi K, Biswas SB. Interaction of extracellular domain 2 of the human retina-specific ATP-binding cassette transporter (ABCA4) with all-trans-retinal. J Biol Chem. 2010; 285:19372-83. [PMID: 20404325]
  4. Xie T, Zhang Z, Fang Q, Du B, Gong X. Structural basis of substrate recognition and translocation by human ABCA4. Nat Commun. 2021; 12:3853 [PMID: 34158497]
  5. Amberger JS, Bocchini CA, Schiettecatte F, Scott AF, Hamosh A. Online Mendelian Inheritance in Man (OMIM®), an online catalog of human genes and genetic disorders. Nucleic Acids Res. 2015; 43:D789-98. [PMID: 25428349]
  6. Stenson PD, Mort M, Ball EV, Shaw K, Phillips A, Cooper DN. The Human Gene Mutation Database: building a comprehensive mutation repository for clinical and molecular genetics, diagnostic testing and personalized genomic medicine. Hum Genet. 2014; 133:1-9. [PMID: 24077912]
  7. Fokkema IF, Taschner PE, Schaafsma GC, Celli J, Laros JF, den Dunnen JT. LOVD v.2.0: the next generation in gene variant databases. Hum Mutat. 2011; 32:557-63. [PMID: 21520333]
  8. Richards S, Aziz N, Bale S, Bick D, Das S, Gastier-Foster J, Grody WW, Hegde M, Lyon E, Spector E, Voelkerding K, Rehm HL, ACMG Laboratory Quality Assurance Committee. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet Med. 2015; 17:405-24. [PMID: 25741868]
  9. Holtan JP, Aukrust I, Jansson RW, Berland S, Bruland O, Gjerde BL, Stokowy T, Bojovic O, Forsaa V, Austeng D, Rødahl E, Bredrup C, Knappskog PM, Bragadóttir R. Clinical features and molecular genetics of patients with ABCA4-retinal dystrophies. Acta Ophthalmol. 2021; 99:e733-46. [PMID: 33258285]
  10. Zhang N, Tsybovsky Y, Kolesnikov AV, Rozanowska M, Swider M, Schwartz SB, Stone EM, Palczewska G, Maeda A, Kefalov VJ, Jacobson SG, Cideciyan AV, Palczewski K. Protein misfolding and the pathogenesis of ABCA4-associated retinal degenerations. Hum Mol Genet. 2015; 24:3220-37. [PMID: 25712131]
  11. Shroyer NF, Lewis RA, Yatsenko AN, Wensel TG, Lupski JR. Cosegregation and functional analysis of mutant ABCR (ABCA4) alleles in families that manifest both Stargardt disease and age-related macular degeneration. Hum Mol Genet. 2001; 10:2671-8. [PMID: 11726554]
  12. Cremers FPM, Lee W, Collin RWJ, Allikmets R. Clinical spectrum, genetic complexity and therapeutic approaches for retinal disease caused by ABCA4 mutations. Prog Retin Eye Res. 2020; 79100861 [PMID: 32278709]
  13. Biswas-Fiss EE. Functional analysis of genetic mutations in nucleotide binding domain 2 of the human retina specific ABC transporter. Biochemistry. 2003; 42:10683-96. [PMID: 12962493]
  14. Patel MJ, Biswas SB, Biswas-Fiss EE. Functional significance of the conserved C-Terminal VFVNFA motif in the retina-specific ABC transporter, ABCA4, and its role in inherited visual disease. Biochem Biophys Res Commun. 2019; 519:46-52. [PMID: 31481235]
  15. Suárez T, Biswas SB, Biswas EE. Biochemical defects in retina-specific human ATP binding cassette transporter nucleotide binding domain 1 mutants associated with macular degeneration. J Biol Chem. 2002; 277:21759-67. [PMID: 11919200]
  16. Wangtiraumnuay N, Capasso J, Tsukikawa M, Levin A, Biswas-Fiss E. Novel ABCA4 mutation leads to loss of a conserved C-terminal motif: implications for predicting pathogenicity based on genetic testing. Eur J Ophthalmol. 2018; 28:123-6. [PMID: 28885670]
  17. Garces F, Jiang K, Molday LL, Stöhr H, Weber BH, Lyons CJ, Maberley D, Molday RS. Correlating the Expression and Functional Activity of ABCA4 Disease Variants With the Phenotype of Patients With Stargardt Disease. Invest Ophthalmol Vis Sci. 2018; 59:2305-15. [PMID: 29847635]
  18. Molday LL, Wahl D, Sarunic MV, Molday RS. Localization and functional characterization of the p.Asn965Ser (N965S) ABCA4 variant in mice reveal pathogenic mechanisms underlying Stargardt macular degeneration. Hum Mol Genet. 2018; 27:295-306. [PMID: 29145636]
  19. Garces FA, Scortecci JF, Molday RS. Functional Characterization of ABCA4 Missense Variants Linked to Stargardt Macular Degeneration. Int J Mol Sci. 2020; 22:185 [PMID: 33375396]
  20. Sun H, Smallwood PM, Nathans J. Biochemical defects in ABCR protein variants associated with human retinopathies. Nat Genet. 2000; 26:242-6. [PMID: 11017087]
  21. Lee W, Schuerch K, Zernant J, Collison FT, Bearelly S, Fishman GA, Tsang SH, Sparrow JR, Allikmets R. Genotypic spectrum and phenotype correlations of ABCA4-associated disease in patients of south Asian descent. Eur J Hum Genet. 2017; 25:735-43. [PMID: 28327576]
  22. Westeneng-van Haaften SC, Boon CJF, Cremers FPM, Hoefsloot LH, den Hollander AI, Hoyng CB. Clinical and genetic characteristics of late-onset Stargardt’s disease. Ophthalmology. 2012; 119:1199-210. [PMID: 22449572]
  23. Wiszniewski W, Zaremba CM, Yatsenko AN, Jamrich M, Wensel TG, Lewis RA, Lupski JR. ABCA4 mutations causing mislocalization are found frequently in patients with severe retinal dystrophies. Hum Mol Genet. 2005; 14:2769-78. [PMID: 16103129]
  24. Zernant J, Lee W, Nagasaki T, Collison FT, Fishman GA, Bertelsen M, Rosenberg T, Gouras P, Tsang SH, Allikmets R. Extremely hypomorphic and severe deep intronic variants in the ABCA4 locus result in varying Stargardt disease phenotypes. Cold Spring Harb Mol Case Stud. 2018; 4a002733 [PMID: 29848554]
  25. Ittisoponpisan S, Islam SA, Khanna T, Alhuzimi E, David A, Sternberg MJE. Can Predicted Protein 3D Structures Provide Reliable Insights into whether Missense Variants Are Disease Associated? J Mol Biol. 2019; 431:2197-212. [PMID: 30995449]
  26. Kalay E, Uzumcu A, Krieger E, Çaylan R, Uyguner O, Ulubil-Emiroglu M, Erdol H, Kayserili H, Hafiz G, Başerer N, Heister AJGM, Hennies HC, Nürnberg P, Başaran S, Brunner HG, Cremers CWRJ, Karaguzel A, Wollnik B, Kremer H. MYO15A (DFNB3) mutations in Turkish hearing loss families and functional modeling of a novel motor domain mutation. Am J Med Genet A. 2007; 143A:2382-9. [PMID: 17853461]
  27. Kueppers F, Andrake MD, Xu Q, Dunbrack RL, , Jr Kim J, Sanders CL. Protein modeling to assess the pathogenicity of rare variants of SERPINA1 in patients suspected of having Alpha 1 Antitrypsin Deficiency. BMC Med Genet. 2019; 20:125 [PMID: 31307431]
  28. Ou L, Przybilla MJ, Whitley CB. Phenotype prediction for mucopolysaccharidosis type I by in silico analysis. Orphanet J Rare Dis. 2017; 12:125 [PMID: 28676128]
  29. Sallah SR, Sergouniotis PI, Barton S, Ramsden S, Taylor RL, Safadi A, Kabir M, Ellingford JM, Lench N, Lovell SC, Black GCM. Using an integrative machine learning approach utilising homology modelling to clinically interpret genetic variants: CACNA1F as an exemplar. Eur J Hum Genet. 2020; 28:1274-82. [PMID: 32313206]
  30. Rhoades R, Henry B, Prichett D, Fang Y, Teng S. Computational Saturation Mutagenesis to Investigate the Effects of Neurexin-1 Mutations on AlphaFold Structure. Genes (Basel). 2022; 13:789 [PMID: 35627176]
  31. Buonfiglio PI, Bruque CD, Lotersztein V, Luce L, Giliberto F, Menazzi S, Francipane L, Paoli B, Goldschmidt E, Elgoyhen AB, Dalamón V. Predicting pathogenicity for novel hearing loss mutations based on genetic and protein structure approaches. Sci Rep. 2022; 12:301 [PMID: 34997062]
  32. Bruque CD, Delea M, Fernández CS, Orza JV, Taboas M, Buzzalino N, Espeche LD, Solari A, Luccerini V, Alba L, Nadra AD, Dain L. Structure-based activity prediction of CYP21A2 stability variants: A survey of available gene variations. Sci Rep. 2016; 6:39082 [PMID: 27966633]
  33. Caswell RC, Owens MM, Gunning AC, Ellard S, Wright CF. Using Structural Analysis In Silico to Assess the Impact of Missense Variants in MEN1. J Endocr Soc. 2019; 3:2258-75. [PMID: 31737856]
  34. Cevik S, Biswas SB, Biswas-Fiss EE. Structural and Pathogenic Impacts of ABCA4 Variants in Retinal Degenerations-An In-Silico Study. Int J Mol Sci. 2023; 24:7280 [PMID: 37108442]
  35. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, Tunyasuvunakool K, Bates R, Žídek A, Potapenko A, Bridgland A, Meyer C, Kohl SAA, Ballard AJ, Cowie A, Romera-Paredes B, Nikolov S, Jain R, Adler J, Back T, Petersen S, Reiman D, Clancy E, Zielinski M, Steinegger M, Pacholska M, Berghammer T, Bodenstein S, Silver D, Vinyals O, Senior AW, Kavukcuoglu K, Kohli P, Hassabis D. Highly accurate protein structure prediction with AlphaFold. Nature. 2021; 596:583-9. [PMID: 34265844]
  36. McCulloch DL, Marmor MF, Brigell MG, Hamilton R, Holder GE, Tzekov R, Bach M. ISCEV Standard for full-field clinical electroretinography (2015 update). Doc Ophthalmol. 2015; 130:1-12. [PMID: 25502644]
  37. Liu F, Lee J, Chen J. Molecular structures of the eukaryotic retinal importer ABCA4. Elife. 2021; 10:10 [PMID: 33605212]
  38. Scortecci JF, Molday LL, Curtis SB, Garces FA, Panwar P, Van Petegem F, Molday RS. Cryo-EM structures of the ABCA4 importer reveal mechanisms underlying substrate binding and Stargardt disease. Nat Commun. 2021; 12:5902 [PMID: 34625547]
  39. Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S, Steinegger M. ColabFold: making protein folding accessible to all. Nat Methods. 2022; 19:679-82. [PMID: 35637307]
  40. Krieger E, Vriend G. YASARA View - molecular graphics for all devices - from smartphones to workstations. Bioinformatics. 2014; 30:2981-2. [PMID: 24996895]
  41. Schymkowitz J, Borg J, Stricher F, Nys R, Rousseau F, Serrano L. The FoldX web server: an online force field. Nucleic Acids Res. 2005; 33suppl_2W382-8 [PMID: 15980494]
  42. Rentzsch P, Witten D, Cooper GM, Shendure J, Kircher M. CADD: predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Res. 2019; 47D1:D886-94. [PMID: 30371827]
  43. Adzhubei IA, Schmidt S, Peshkin L, Ramensky VE, Gerasimova A, Bork P, Kondrashov AS, Sunyaev SR. A method and server for predicting damaging missense mutations. Nat Methods. 2010; 7:248-9. [PMID: 20354512]
  44. Ioannidis NM, Rothstein JH, Pejaver V, Middha S, McDonnell SK, Baheti S, Musolf A, Li Q, Holzinger E, Karyadi D, Cannon-Albright LA, Teerlink CC, Stanford JL, Isaacs WB, Xu J, Cooney KA, Lange EM, Schleutker J, Carpten JD, Powell IJ, Cussenot O, Cancel-Tassin G, Giles GG, MacInnis RJ, Maier C, Hsieh C-L, Wiklund F, Catalona WJ, Foulkes WD, Mandal D, Eeles RA, Kote-Jarai Z, Bustamante CD, Schaid DJ, Hastie T, Ostrander EA, Bailey-Wilson JE, Radivojac P, Thibodeau SN, Whittemore AS, Sieh W. REVEL: An Ensemble Method for Predicting the Pathogenicity of Rare Missense Variants. Am J Hum Genet. 2016; 99:877-85. [PMID: 27666373]
  45. Ashkenazy H, Abadi S, Martz E, Chay O, Mayrose I, Pupko T, Ben-Tal N. ConSurf 2016: an improved methodology to estimate and visualize evolutionary conservation in macromolecules. Nucleic Acids Res. 2016; 44W1W344-50 [PMID: 27166375]
  46. Desmet FO, Hamroun D, Lalande M, Collod-Béroud G, Claustres M, Béroud C. Human Splicing Finder: an online bioinformatics tool to predict splicing signals. Nucleic Acids Res. 2009; 37e67 [PMID: 19339519]
  47. Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alföldi J, Wang Q, Collins RL, Laricchia KM, Ganna A, Birnbaum DP, Gauthier LD, Brand H, Solomonson M, Watts NA, Rhodes D, Singer-Berk M, England EM, Seaby EG, Kosmicki JA, Walters RK, Tashman K, Farjoun Y, Banks E, Poterba T, Wang A, Seed C, Whiffin N, Chong JX, Samocha KE, Pierce-Hoffman E, Zappala Z, O’Donnell-Luria AH, Minikel EV, Weisburd B, Lek M, Ware JS, Vittal C, Armean IM, Bergelson L, Cibulskis K, Connolly KM, Covarrubias M, Donnelly S, Ferriera S, Gabriel S, Gentry J, Gupta N, Jeandet T, Kaplan D, Llanwarne C, Munshi R, Novod S, Petrillo N, Roazen D, Ruano-Rubio V, Saltzman A, Schleicher M, Soto J, Tibbetts K, Tolonen C, Wade G, Talkowski ME, Neale BM, Daly MJ, MacArthur DG, Ardissino D, Atzmon G, Barnard J, Beaugerie L, Benjamin EJ, Boehnke M, Bonnycastle LL, Bottinger EP, Bowden DW, Bown MJ, Chambers JC, Chan JC, Chasman D, Cho J, Chung MK, Cohen B, Correa A, Dabelea D, Daly MJ, Darbar D, Duggirala R, Dupuis J, Ellinor PT, Elosua R, Erdmann J, Esko T, Färkkilä M, Florez J, Franke A, Getz G, Glaser B, Glatt SJ, Goldstein D, Gonzalez C, Groop L, Haiman C, Hanis C, Harms M, Hiltunen M, Holi MM, Hultman CM, Kallela M, Kaprio J, Kathiresan S, Kim B-J, Kim YJ, Kirov G, Kooner J, Koskinen S, Krumholz HM, Kugathasan S, Kwak SH, Laakso M, Lehtimäki T, Loos RJF, Lubitz SA, Ma RCW, MacArthur DG, Marrugat J, Mattila KM, McCarroll S, McCarthy MI, McGovern D, McPherson R, Meigs JB, Melander O, Metspalu A, Neale BM, Nilsson PM, O’Donovan MC, Ongur D, Orozco L, Owen MJ, Palmer CNA, Palotie A, Park KS, Pato C, Pulver AE, Rahman N, Remes AM, Rioux JD, Ripatti S, Roden DM, Saleheen D, Salomaa V, Samani NJ, Scharf J, Schunkert H, Shoemaker MB, Sklar P, Soininen H, Sokol H, Spector T, Sullivan PF, Suvisaari J, Tai ES, Teo YY, Tiinamaija T, Tsuang M, Turner D, Tusie-Luna T, Vartiainen E, Vawter MP, Ware JS, Watkins H, Weersma RK, Wessman M, Wilson JG, Xavier RJ, Neale BM, Daly MJ, MacArthur DG, Genome Aggregation Database C, Genome Aggregation Database Consortium. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature. 2020; 581:434-43. [PMID: 32461654]
  48. Albert S, Garanto A, Sangermano R, Khan M, Bax NM, Hoyng CB, Zernant J, Lee W, Allikmets R, Collin RWJ, Cremers FPM. Identification and Rescue of Splice Defects Caused by Two Neighboring Deep-Intronic ABCA4 Mutations Underlying Stargardt Disease. Am J Hum Genet. 2018; 102:517-27. [PMID: 29526278]
  49. Al-Khuzaei S, Broadgate S, Foster CR, Shah M, Yu J, Downes SM, Halford S. An Overview of the Genetics of ABCA4 Retinopathies, an Evolving Story. Genes (Basel). 2021; 12:1241 [PMID: 34440414]
  50. Khan M, Cremers FPM. ABCA4-Associated Stargardt Disease. Klin Monbl Augenheilkd. 2020; 237:267-74. [PMID: 32016942]
  51. Fowler NJ, Williamson MP. The accuracy of protein structures in solution determined by AlphaFold and NMR. Structure. 2022; 30:925-933.e2. [PMID: 35537451]
  52. Akdel M, Pires DEV, Pardo EP, Jänes J, Zalevsky AO, Mészáros B, Bryant P, Good LL, Laskowski RA, Pozzati G, Shenoy A, Zhu W, Kundrotas P, Serra VR, Rodrigues CHM, Dunham AS, Burke D, Borkakoti N, Velankar S, Frost A, Basquin J, Lindorff-Larsen K, Bateman A, Kajava AV, Valencia A, Ovchinnikov S, Durairaj J, Ascher DB, Thornton JM, Davey NE, Stein A, Elofsson A, Croll TI, Beltrao P. A structural biology community assessment of AlphaFold2 applications. Nat Struct Mol Biol. 2022; 29:1056-67. [PMID: 36344848]
  53. Prabantu VM, Naveenkumar N, Srinivasan N. Influence of Disease-Causing Mutations on Protein Structural Networks. Front Mol Biosci. 2021; 7620554 [PMID: 33778000]
  54. Curtis SB, Molday LL, Garces FA, Molday RS. Functional analysis and classification of homozygous and hypomorphic ABCA4 variants associated with Stargardt macular degeneration. Hum Mutat. 2020; 41:1944-56. [PMID: 32845050]
  55. Jones PM, George AM. Mechanism of ABC transporters: a molecular dynamics simulation of a well characterized nucleotide-binding subunit. Proc Natl Acad Sci U S A. 2002; 99:12639-44. [PMID: 12237398]
  56. Khalili-Araghi F, Gumbart J, Wen P-C, Sotomayor M, Tajkhorshid E, Schulten K. Molecular dynamics simulations of membrane channels and transporters. Curr Opin Struct Biol. 2009; 19:128-37. [PMID: 19345092]