Performance Evaluation of SpliceAI for the Prediction of Splicing of NF1 Variants

Investigation of cryptic JAG1 splice variants as a cause of Alagille syndrome and performance evaluation of splice predictor tools

Haploinsufficiency of JAG1 is the primary cause of Alagille syndrome (ALGS), a rare, multisystem disorder. The identification of JAG1 intronic variants outside of the canonical splice region as well as missense variants, both of which lead to uncertain associations with disease, confuses diagnostics. Strategies to determine whether these variants affect splicing include the study of patient RNA or minigene constructs, which are not always available or can be laborious to design, as well as the utilization of computational splice prediction tools. These tools, including SpliceAI and Pangolin, use algorithms to calculate the probability that a variant results in a splice alteration, expressed as a Δ score, with higher Δ scores (>0.2 on a 0-1 scale) positively correlated with aberrant splicing. We studied the consequence of 10 putative splice variants in ALGS patient samples through RNA analysis and compared this to SpliceAI and Pangolin predictions. We identified eight variants with aberrant splicing, seven of which had not been previously validated. Combining these data with non-canonical and missense splice variants reported in the literature, we identified a predictive threshold for SpliceAI and Pangolin with high sensitivity (Δ score >0.6). Moreover, we showed reduced specificity for variants with low Δ scores (<0.2), highlighting a limitation of these tools that results in the misidentification of true splice variants. These results improve genomic diagnostics for ALGS by confirming splice effects for seven variants and suggest that the integration of splice prediction tools with RNA analysis is important to ensure accurate clinical variant classifications.

Functional evaluation of rare variants in complement factor I using a minigene assay

The regulatory serine protease, complement factor I (FI), in conjunction with one of its cofactors (FH, C4BP, MCP, or CR1), plays an essential role in controlling complement activity through inactivation of C3b and C4b. The functional impact by missense variants in the CFI gene, particularly those with minor allele frequencies of 0.01% to 0.1%, is infrequently studied. As such, these variants are typically classified as variants of uncertain significance (VUS) when they are identified by clinical testing. Herein, we utilized a minigene splicing assay to assess the functional impact of 36 ultra-rare variants of CFI. These variants were selected based on their minor allele frequencies (MAF) and their association with low-normal FI levels. Four variants lead to aberrant splicing-one 5' consensus splice site (NM_000204.5: c.1429G>C, p.Asp477His) and three exonic changes (c.355G>A, p.Gly119Arg; c.472G>A, p.Gly158Arg; and c.950G>A, p.Arg317Gln)-enabling their reclassification to likely pathogenic (LP) or pathogenic (P) based on ACMG guidelines. These findings underscore the value of functional assays, such as the minigene assay, in assessing the clinical relevance of rare variants in CFI.

Genotype-Phenotype Correlation in Junctional Epidermolysis Bullosa: Signposts to Severity

Junctional epidermolysis bullosa (JEB) is a rare autosomal recessive genodermatosis with a broad spectrum of phenotypes. Current genotype-phenotype paradigms are insufficient to accurately predict JEB subtype and characteristics from genotype, particularly for splice site variants, which account for over a fifth of disease-causing variants in JEB. This study evaluated the genetic and clinical findings from a JEB cohort, investigating genotype-phenotype correlations through bioinformatic analyses and comparison with previously reported variants. Eighteen unique variants in LAMB3, LAMA3, LAMC2, or COL17A1 were identified from 17 individuals. Seven had severe JEB, 9 had intermediate JEB, and 1 had laryngo-onycho-cutaneous syndrome. Seven variants were previously unreported. Deep phenotyping was completed for all intermediate JEB cases and demonstrated substantial variation between individuals. Splice site variants underwent analysis with SpliceAI, a state-of-the-art artificial intelligence tool, to predict resultant transcripts. Predicted functional effects included exon skipping and cryptic splice site activation, which provided potential explanations for disease severity and in most cases correlated with laminin-332 immunofluorescence. RT-PCR was performed for 1 case to investigate resultant transcripts produced from the splice site variant. This study expands the JEB genomic and phenotypic landscape. Artificial intelligence tools show potential for predicting the functional effects of splice site variants and may identify candidates for confirmatory laboratory investigation. Investigation of RNA transcripts will help to further elucidate genotype-phenotype correlations for novel variants.

Predicting the impact of rare variants on RNA splicing in CAGI6

Variants which disrupt splicing are a frequent cause of rare disease that have been under-ascertained clinically. Accurate and efficient methods to predict a variant's impact on splicing are needed to interpret the growing number of variants of unknown significance (VUS) identified by exome and genome sequencing. Here, we present the results of the CAGI6 Splicing VUS challenge, which invited predictions of the splicing impact of 56 variants ascertained clinically and functionally validated to determine splicing impact. The performance of 12 prediction methods, along with SpliceAI and CADD, was compared on the 56 functionally validated variants. The maximum accuracy achieved was 82% from two different approaches, one weighting SpliceAI scores by minor allele frequency, and one applying the recently published Splicing Prediction Pipeline (SPiP). SPiP performed optimally in terms of sensitivity, while an ensemble method combining multiple prediction tools and information from databases exceeded all others for specificity. Several challenge methods equalled or exceeded the performance of SpliceAI, with ultimate choice of prediction method likely to depend on experimental or clinical aims. One quarter of the variants were incorrectly predicted by at least 50% of the methods, highlighting the need for further improvements to splicing prediction methods for successful clinical application.

Benchmarking splice variant prediction algorithms using massively parallel splicing assays

BACKGROUND

Variants that disrupt mRNA splicing account for a sizable fraction of the pathogenic burden in many genetic disorders, but identifying splice-disruptive variants (SDVs) beyond the essential splice site dinucleotides remains difficult. Computational predictors are often discordant, compounding the challenge of variant interpretation. Because they are primarily validated using clinical variant sets heavily biased to known canonical splice site mutations, it remains unclear how well their performance generalizes.

RESULTS

We benchmark eight widely used splicing effect prediction algorithms, leveraging massively parallel splicing assays (MPSAs) as a source of experimentally determined ground-truth. MPSAs simultaneously assay many variants to nominate candidate SDVs. We compare experimentally measured splicing outcomes with bioinformatic predictions for 3,616 variants in five genes. Algorithms' concordance with MPSA measurements, and with each other, is lower for exonic than intronic variants, underscoring the difficulty of identifying missense or synonymous SDVs. Deep learning-based predictors trained on gene model annotations achieve the best overall performance at distinguishing disruptive and neutral variants, and controlling for overall call rate genome-wide, SpliceAI and Pangolin have superior sensitivity. Finally, our results highlight two practical considerations when scoring variants genome-wide: finding an optimal score cutoff, and the substantial variability introduced by differences in gene model annotation, and we suggest strategies for optimal splice effect prediction in the face of these issues.

CONCLUSION

SpliceAI and Pangolin show the best overall performance among predictors tested, however, improvements in splice effect prediction are still needed especially within exons.

Intronic Germline DICER1 Variants in Patients With Sertoli-Leydig Cell Tumor

Germline pathogenic loss-of-function (pLOF) variants in DICER1 are associated with a predisposition for a variety of solid neoplasms, including pleuropulmonary blastoma and Sertoli-Leydig cell tumor (SLCT). The most common DICER1 pLOF variants include small insertions or deletions leading to frameshifts, and base substitutions leading to nonsense codons or altered splice sites. Larger deletions and pathogenic missense variants occur less frequently. Identifying these variants can trigger surveillance algorithms with potential for early detection of DICER1-related cancers and cascade testing of family members. However, some patients with DICER1-associated tumors have no pLOF variants detected by germline or tumor testing. Here, we present two patients with SLCT whose tumor sequencing showed only a somatic missense DICER1 RNase IIIb variant. Conventional exon-directed germline sequencing revealed no pLOF variants. Using a custom capture panel, we discovered novel intronic variants, ENST00000343455.7: c.1752+213A>G and c.1509+16A>G, that appear to interfere with normal splicing. We suggest that when no DICER1 pLOF variants or large deletions are discovered in exonic regions despite strong clinical suspicion, intron sequencing and splicing analysis should be performed.

DECIPHER: Improving Genetic Diagnosis Through Dynamic Integration of Genomic and Clinical Data

DECIPHER (Database of Genomic Variation and Phenotype in Humans Using Ensembl Resources) shares candidate diagnostic variants and phenotypic data from patients with genetic disorders to facilitate research and improve the diagnosis, management, and therapy of rare diseases. The platform sits at the boundary between genomic research and the clinical community. DECIPHER aims to ensure that the most up-to-date data are made rapidly available within its interpretation interfaces to improve clinical care. Newly integrated cardiac case-control data that provide evidence of gene-disease associations and inform variant interpretation exemplify this mission. New research resources are presented in a format optimized for use by a broad range of professionals supporting the delivery of genomic medicine. The interfaces within DECIPHER integrate and contextualize variant and phenotypic data, helping to determine a robust clinico-molecular diagnosis for rare-disease patients, which combines both variant classification and clinical fit. DECIPHER supports discovery research, connecting individuals within the rare-disease community to pursue hypothesis-driven research.

Using the ACMG/AMP framework to capture evidence related to predicted and observed impact on splicing: Recommendations from the ClinGen SVI Splicing Subgroup

The American College of Medical Genetics and Genomics (ACMG)/Association for Molecular Pathology (AMP) framework for classifying variants uses six evidence categories related to the splicing potential of variants: PVS1, PS3, PP3, BS3, BP4, and BP7. However, the lack of guidance on how to apply such codes has contributed to variation in the specifications developed by different Clinical Genome Resource (ClinGen) Variant Curation Expert Panels. The ClinGen Sequence Variant Interpretation Splicing Subgroup was established to refine recommendations for applying ACMG/AMP codes relating to splicing data and computational predictions. We utilized empirically derived splicing evidence to (1) determine the evidence weighting of splicing-related data and appropriate criteria code selection for general use, (2) outline a process for integrating splicing-related considerations when developing a gene-specific PVS1 decision tree, and (3) exemplify methodology to calibrate splice prediction tools. We propose repurposing the PVS1_Strength code to capture splicing assay data that provide experimental evidence for variants resulting in RNA transcript(s) with loss of function. Conversely, BP7 may be used to capture RNA results demonstrating no splicing impact for intronic and synonymous variants. We propose that the PS3/BS3 codes are applied only for well-established assays that measure functional impact not directly captured by RNA-splicing assays. We recommend the application of PS1 based on similarity of predicted RNA-splicing effects for a variant under assessment in comparison with a known pathogenic variant. The recommendations and approaches for consideration and evaluation of RNA-assay evidence described aim to help standardize variant pathogenicity classification processes when interpreting splicing-based evidence.

Benchmarking splice variant prediction algorithms using massively parallel splicing assays

Background

Variants that disrupt mRNA splicing account for a sizable fraction of the pathogenic burden in many genetic disorders, but identifying splice-disruptive variants (SDVs) beyond the essential splice site dinucleotides remains difficult. Computational predictors are often discordant, compounding the challenge of variant interpretation. Because they are primarily validated using clinical variant sets heavily biased to known canonical splice site mutations, it remains unclear how well their performance generalizes.

Results

We benchmarked eight widely used splicing effect prediction algorithms, leveraging massively parallel splicing assays (MPSAs) as a source of experimentally determined ground-truth. MPSAs simultaneously assay many variants to nominate candidate SDVs. We compared experimentally measured splicing outcomes with bioinformatic predictions for 3,616 variants in five genes. Algorithms' concordance with MPSA measurements, and with each other, was lower for exonic than intronic variants, underscoring the difficulty of identifying missense or synonymous SDVs. Deep learning-based predictors trained on gene model annotations achieved the best overall performance at distinguishing disruptive and neutral variants. Controlling for overall call rate genome-wide, SpliceAI and Pangolin also showed superior overall sensitivity for identifying SDVs. Finally, our results highlight two practical considerations when scoring variants genome-wide: finding an optimal score cutoff, and the substantial variability introduced by differences in gene model annotation, and we suggest strategies for optimal splice effect prediction in the face of these issues.

Conclusion

SpliceAI and Pangolin showed the best overall performance among predictors tested, however, improvements in splice effect prediction are still needed especially within exons.

APPLICATION OF THE ACMG/AMP FRAMEWORK TO CAPTURE EVIDENCE RELEVANT TO PREDICTED AND OBSERVED IMPACT ON SPLICING: RECOMMENDATIONS FROM THE CLINGEN SVI SPLICING SUBGROUP

The American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP) framework for classifying variants uses six evidence categories related to the splicing potential of variants: PVS1 (null variant in a gene where loss-of-function is the mechanism of disease), PS3 (functional assays show damaging effect on splicing), PP3 (computational evidence supports a splicing effect), BS3 (functional assays show no damaging effect on splicing), BP4 (computational evidence suggests no splicing impact), and BP7 (silent change with no predicted impact on splicing). However, the lack of guidance on how to apply such codes has contributed to variation in the specifications developed by different Clinical Genome Resource (ClinGen) Variant Curation Expert Panels. The ClinGen Sequence Variant Interpretation (SVI) Splicing Subgroup was established to refine recommendations for applying ACMG/AMP codes relating to splicing data and computational predictions. Our study utilised empirically derived splicing evidence to: 1) determine the evidence weighting of splicing-related data and appropriate criteria code selection for general use, 2) outline a process for integrating splicing-related considerations when developing a gene-specific PVS1 decision tree, and 3) exemplify methodology to calibrate bioinformatic splice prediction tools. We propose repurposing of the PVS1_Strength code to capture splicing assay data that provide experimental evidence for variants resulting in RNA transcript(s) with loss of function. Conversely BP7 may be used to capture RNA results demonstrating no impact on splicing for both intronic and synonymous variants, and for missense variants if protein functional impact has been excluded. Furthermore, we propose that the PS3 and BS3 codes are applied only for well-established assays that measure functional impact that is not directly captured by RNA splicing assays. We recommend the application of PS1 based on similarity of predicted RNA splicing effects for a variant under assessment in comparison to a known Pathogenic variant. The recommendations and approaches for consideration and evaluation of RNA assay evidence described aim to help standardise variant pathogenicity classification processes and result in greater consistency when interpreting splicing-based evidence.

Systematic analysis of CNGA3 splice variants identifies different mechanisms of aberrant splicing

Achromatopsia is an autosomal recessive cone photoreceptor disease that is frequently caused by pathogenic variants in the CNGA3 gene. Here, we present a systematic functional analysis of 20 CNGA3 splice site variants detected in our large cohort of achromatopsia patients and/or listed in common variant databases. All variants were analyzed by functional splice assays based on the pSPL3 exon trapping vector. We demonstrated that ten variants, both at canonical and non-canonical splice sites, induced aberrant splicing, including intronic nucleotide retention, exonic nucleotide deletion and exon skipping, resulting in 21 different aberrant transcripts. Of these, eleven were predicted to introduce a premature termination codon. The pathogenicity of all variants was assessed based on established guidelines for variant classification. Incorporation of the results of our functional analyses enabled re-classification of 75% of variants previously classified as variants of uncertain significance into either likely benign or likely pathogenic. Our study is the first in which a systematic characterization of putative CNGA3 splice variants has been performed. We demonstrated the utility of pSPL3 based minigene assays in the effective assessment of putative splice variants. Our findings improve the diagnosis of achromatopsia patients, who may thus benefit from future gene-based therapeutic strategies.

SpliceAI-visual: a free online tool to improve SpliceAI splicing variant interpretation

SpliceAI is an open-source deep learning splicing prediction algorithm that has demonstrated in the past few years its high ability to predict splicing defects caused by DNA variations. However, its outputs present several drawbacks: (1) although the numerical values are very convenient for batch filtering, their precise interpretation can be difficult, (2) the outputs are delta scores which can sometimes mask a severe consequence, and (3) complex delins are most often not handled. We present here SpliceAI-visual, a free online tool based on the SpliceAI algorithm, and show how it complements the traditional SpliceAI analysis. First, SpliceAI-visual manipulates raw scores and not delta scores, as the latter can be misleading in certain circumstances. Second, the outcome of SpliceAI-visual is user-friendly thanks to the graphical presentation. Third, SpliceAI-visual is currently one of the only SpliceAI-derived implementations able to annotate complex variants (e.g., complex delins). We report here the benefits of using SpliceAI-visual and demonstrate its relevance in the assessment/modulation of the PVS1 classification criteria. We also show how SpliceAI-visual can elucidate several complex splicing defects taken from the literature but also from unpublished cases. SpliceAI-visual is available as a Google Colab notebook and has also been fully integrated in a free online variant interpretation tool, MobiDetails ( https://mobidetails.iurc.montp.inserm.fr/MD ).

Genetic analyses of DNA repair pathway associated genes implicate new candidate cancer predisposing genes in ancestrally defined ovarian cancer cases

Not all familial ovarian cancer (OC) cases are explained by pathogenic germline variants in known risk genes. A candidate gene approach involving DNA repair pathway genes was applied to identify rare recurring pathogenic variants in familial OC cases not associated with known OC risk genes from a population exhibiting genetic drift. Whole exome sequencing (WES) data of 15 OC cases from 13 families tested negative for pathogenic variants in known OC risk genes were investigated for candidate variants in 468 DNA repair pathway genes. Filtering and prioritization criteria were applied to WES data to select top candidates for further analyses. Candidates were genotyped in ancestry defined study groups of 214 familial and 998 sporadic OC or breast cancer (BC) cases and 1025 population-matched controls and screened for additional carriers in 605 population-matched OC cases. The candidate genes were also analyzed in WES data from 937 familial or sporadic OC cases of diverse ancestries. Top candidate variants in ERCC5, EXO1, FANCC, NEIL1 and NTHL1 were identified in 5/13 (39%) OC families. Collectively, candidate variants were identified in 7/435 (1.6%) sporadic OC cases and 1/566 (0.2%) sporadic BC cases versus 1/1025 (0.1%) controls. Additional carriers were identified in 6/605 (0.9%) OC cases. Tumour DNA from ERCC5, NEIL1 and NTHL1 variant carriers exhibited loss of the wild-type allele. Carriers of various candidate variants in these genes were identified in 31/937 (3.3%) OC cases of diverse ancestries versus 0-0.004% in cancer-free controls. The strategy of applying a candidate gene approach in a population exhibiting genetic drift identified new candidate OC predisposition variants in DNA repair pathway genes.

Aberrant splicing caused by exonic single nucleotide variants positioned 2nd or 3rd to the last nucleotide in the COL4A5 gene

BACKGROUND AND OBJECTIVES

The evident genotype-phenotype correlation shown by the X-linked Alport syndrome warrants the assessment of the impact of identified gene variants on aberrant splicing. We previously reported that single nucleotide variants (SNVs) in the last nucleotide of exons in COL4A5 cause aberrant splicing. It is known that the nucleotides located 2nd and 3rd to the last nucleotides of exons can also play an essential role in the first step of the splicing process. In this study, we aimed to investigate whether SNVs positioned 2nd or 3rd to the last nucleotide of exons in COL4A5 resulted in aberrant splicing.

METHODS

We selected eight candidate variants: six from the Human Gene Variant Database Professional and two from our cohort. We performed an in-vitro splicing assay and reverse transcription-polymerase chain reaction (RT-PCR) for messenger RNA obtained from patients, if available.

RESULTS

The candidate variants were initially classified into the following groups: three nonsense, two missense, and three synonymous variants. Splicing assays and RT-PCR for messenger RNA revealed that six of the eight variants caused aberrant splicing. Four variants, initially classified as non-truncating variants, were found to be truncating ones, which usually show relatively more severe phenotypes.

CONCLUSION

We revealed that exonic SNVs positioned 2nd or 3rd to the last nucleotide of exons in the COL4A5 were responsible for aberrant splicing. The results of our study suggest that attention should be paid when interpreting the pathogenicity of exonic SNVs near the 5' splice site.

Computational prediction of human deep intronic variation

BACKGROUND

The adoption of whole-genome sequencing in genetic screens has facilitated the detection of genetic variation in the intronic regions of genes, far from annotated splice sites. However, selecting an appropriate computational tool to discriminate functionally relevant genetic variants from those with no effect is challenging, particularly for deep intronic regions where independent benchmarks are scarce.

RESULTS

In this study, we have provided an overview of the computational methods available and the extent to which they can be used to analyze deep intronic variation. We leveraged diverse datasets to extensively evaluate tool performance across different intronic regions, distinguishing between variants that are expected to disrupt splicing through different molecular mechanisms. Notably, we compared the performance of SpliceAI, a widely used sequence-based deep learning model, with that of more recent methods that extend its original implementation. We observed considerable differences in tool performance depending on the region considered, with variants generating cryptic splice sites being better predicted than those that potentially affect splicing regulatory elements. Finally, we devised a novel quantitative assessment of tool interpretability and found that tools providing mechanistic explanations of their predictions are often correct with respect to the ground - information, but the use of these tools results in decreased predictive power when compared to black box methods.

CONCLUSIONS

Our findings translate into practical recommendations for tool usage and provide a reference framework for applying prediction tools in deep intronic regions, enabling more informed decision-making by practitioners.

Genetic characterization of 1210 Japanese pedigrees with inherited retinal diseases by whole-exome sequencing

Inherited retinal diseases (IRDs) comprise a phenotypically and genetically heterogeneous group of ocular disorders that cause visual loss via progressive retinal degeneration. Here, we report the genetic characterization of 1210 IRD pedigrees enrolled through the Japan Eye Genetic Consortium and analyzed by whole exome sequencing. The most common phenotype was retinitis pigmentosa (RP, 43%), followed by macular dystrophy/cone- or cone-rod dystrophy (MD/CORD, 13%). In total, 67 causal genes were identified in 37% (448/1210) of the pedigrees. The first and second most frequently mutated genes were EYS and RP1, associated primarily with autosomal recessive (ar) RP, and RP and arMD/CORD, respectively. Examinations of variant frequency in total and by phenotype showed high accountability of a frequent EYS missense variant (c.2528G>A). In addition to the two known EYS founder mutations (c.4957dupA and c.8805C>G) of arRP, we observed a frequent RP1 variant (c.5797C>T) in patients with arMD/CORD.

Comparison of In Silico Tools for Splice-Altering Variant Prediction Using Established Spliceogenic Variants: An End-User’s Point of View

Assessing the impact of variants of unknown significance on splicing has become a critical issue and a bottleneck, especially with the widespread implementation of whole-genome or exome sequencing. Although multiple in silico tools are available, the interpretation and application of these tools are difficult and practical guidelines are still lacking. A streamlined decision-making process can facilitate the downstream RNA analysis in a more efficient manner. Therefore, we evaluated the performance of 8 in silico tools (Splice Site Finder, MaxEntScan, Splice-site prediction by neural network, GeneSplicer, Human Splicing Finder, SpliceAI, Splicing Predictions in Consensus Elements, and SpliceRover) using 114 NF1 spliceogenic variants, experimentally validated at the mRNA level. The change in the predicted score incurred by the variant of the nearest wild-type splice site was analyzed, and for type II, III, and IV splice variants, the change in the prediction score of de novo or cryptic splice site was also analyzed. SpliceAI and SpliceRover, tools based on deep learning, outperformed all other tools, with AUCs of 0.972 and 0.924, respectively. For de novo and cryptic splice sites, SpliceAI outperformed all other tools and showed a sensitivity of 95.7% at an optimal cut-off of 0.02 score change. Our results show that deep learning algorithms, especially those of SpliceAI, are validated at a significantly higher rate than other in silico tools for clinically relevant NF1 variants. This suggests that deep learning algorithms outperform traditional probabilistic approaches and classical machine learning tools in predicting the de novo and cryptic splice sites.

Alternative Splicing in Human Physiology and Disease

Since the discovery of alternative splicing in the late 1970s, a great number of alternatively spliced transcripts have emerged; this number has exponentially increased with the advances in transcriptomics and massive parallel sequencing technologies [...].

Minigene-based splicing analysis and ACMG/AMP-based tentative classification of 56 ATM variants

The ataxia telangiectasia-mutated (ATM) protein is a major coordinator of the DNA damage response pathway. ATM loss-of-function variants are associated with 2-fold increased breast cancer risk. We aimed at identifying and classifying spliceogenic ATM variants detected in subjects of the large-scale sequencing project BRIDGES. A total of 381 variants at the intron-exon boundaries were identified, 128 of which were predicted to be spliceogenic. After further filtering, we ended up selecting 56 variants for splicing analysis. Four functional minigenes (mgATM) spanning exons 4-9, 11-17, 25-29, and 49-52 were constructed in the splicing plasmid pSAD. Selected variants were genetically engineered into the four constructs and assayed in MCF-7/HeLa cells. Forty-eight variants (85.7%) impaired splicing, 32 of which did not show any trace of the full-length (FL) transcript. A total of 43 transcripts were identified where the most prevalent event was exon/multi-exon skipping. Twenty-seven transcripts were predicted to truncate the ATM protein. A tentative ACMG/AMP (American College of Medical Genetics and Genomics/Association for Molecular Pathology)-based classification scheme that integrates mgATM data allowed us to classify 29 ATM variants as pathogenic/likely pathogenic and seven variants as likely benign. Interestingly, the likely pathogenic variant c.1898+2T>G generated 13% of the minigene FL-transcript due to the use of a noncanonical GG-5'-splice-site (0.014% of human donor sites). Circumstantial evidence in three ATM variants (leakiness uncovered by our mgATM analysis together with clinical data) provides some support for a dosage-sensitive expression model in which variants producing ≥30% of FL-transcripts would be predicted benign, while variants producing ≤13% of FL-transcripts might be pathogenic. © 2022 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.