MTSplice predicts effects of genetic variants on tissue-specific splicing

RNA variant assessment using transactivation and transdifferentiation

Understanding the impact of splicing and nonsense variants on RNA is crucial for the resolution of variant classification as well as their suitability for precision medicine interventions. This is primarily enabled through RNA studies involving transcriptomics followed by targeted assays using RNA isolated from clinically accessible tissues (CATs) such as blood or skin of affected individuals. Insufficient disease gene expression in CATs does however pose a major barrier to RNA based investigations, which we show is relevant to 1,436 Mendelian disease genes. We term these "silent" Mendelian genes (SMGs), the largest portion (36%) of which are associated with neurological disorders. We developed two approaches to induce SMG expression in human dermal fibroblasts (HDFs) to overcome this limitation, including CRISPR-activation-based gene transactivation and fibroblast-to-neuron transdifferentiation. Initial transactivation screens involving 40 SMGs stimulated our development of a highly multiplexed transactivation system culminating in the 6- to 90,000-fold induction of expression of 20/20 (100%) SMGs tested in HDFs. Transdifferentiation of HDFs directly to neurons led to expression of 193/516 (37.4%) of SMGs implicated in neurological disease. The magnitude and isoform diversity of SMG expression following either transactivation or transdifferentiation was comparable to clinically relevant tissues. We apply transdifferentiation and/or gene transactivation combined with short- and long-read RNA sequencing to investigate the impact that variants in USH2A, SCN1A, DMD, and PAK3 have on RNA using HDFs derived from affected individuals. Transactivation and transdifferentiation represent rapid, scalable functional genomic solutions to investigate variants impacting SMGs in the patient cell and genomic context.

Multiplexed single-cell characterization of alternative polyadenylation regulators

Most mammalian genes have multiple polyA sites, representing a substantial source of transcript diversity regulated by the cleavage and polyadenylation (CPA) machinery. To better understand how these proteins govern polyA site choice, we introduce CPA-Perturb-seq, a multiplexed perturbation screen dataset of 42 CPA regulators with a 3' scRNA-seq readout that enables transcriptome-wide inference of polyA site usage. We develop a framework to detect perturbation-dependent changes in polyadenylation and characterize modules of co-regulated polyA sites. We find groups of intronic polyA sites regulated by distinct components of the nuclear RNA life cycle, including elongation, splicing, termination, and surveillance. We train and validate a deep neural network (APARENT-Perturb) for tandem polyA site usage, delineating a cis-regulatory code that predicts perturbation response and reveals interactions between regulatory complexes. Our work highlights the potential for multiplexed single-cell perturbation screens to further our understanding of post-transcriptional regulation.

Long-read RNA-seq demarcates cis- and trans-directed alternative RNA splicing

Genetic regulation of alternative splicing constitutes an important link between genetic variation and disease. Nonetheless, RNA splicing is regulated by both cis-acting elements and trans-acting splicing factors. Determining splicing events that are directed primarily by the cis- or trans-acting mechanisms will greatly inform our understanding of the genetic basis of disease. Here, we show that long-read RNA-seq, combined with our new method isoLASER, enables a clear segregation of cis- and trans-directed splicing events for individual samples. The genetic linkage of splicing is largely individual-specific, in stark contrast to the tissue-specific pattern of splicing profiles. Analysis of long-read RNA-seq data from human and mouse revealed thousands of cis-directed splicing events susceptible to genetic regulation. We highlight such events in the HLA genes whose analysis was challenging with short-read data. We also highlight novel cis-directed splicing events in Alzheimer's disease-relevant genes such as MAPT and BIN1. Together, the clear demarcation of cis- and trans-directed splicing paves ways for future studies of the genetic basis of disease.

Self-supervised learning on millions of primary RNA sequences from 72 vertebrates improves sequence-based RNA splicing prediction

Language models pretrained by self-supervised learning (SSL) have been widely utilized to study protein sequences, while few models were developed for genomic sequences and were limited to single species. Due to the lack of genomes from different species, these models cannot effectively leverage evolutionary information. In this study, we have developed SpliceBERT, a language model pretrained on primary ribonucleic acids (RNA) sequences from 72 vertebrates by masked language modeling, and applied it to sequence-based modeling of RNA splicing. Pretraining SpliceBERT on diverse species enables effective identification of evolutionarily conserved elements. Meanwhile, the learned hidden states and attention weights can characterize the biological properties of splice sites. As a result, SpliceBERT was shown effective on several downstream tasks: zero-shot prediction of variant effects on splicing, prediction of branchpoints in humans, and cross-species prediction of splice sites. Our study highlighted the importance of pretraining genomic language models on a diverse range of species and suggested that SSL is a promising approach to enhance our understanding of the regulatory logic underlying genomic sequences.

Evaluating the representational power of pre-trained DNA language models for regulatory genomics

The emergence of genomic language models (gLMs) offers an unsupervised approach to learn a wide diversity of cis-regulatory patterns in the non-coding genome without requiring labels of functional activity generated by wet-lab experiments. Previous evaluations have shown pre-trained gLMs can be leveraged to improve prediction performance across a broad range of regulatory genomics tasks, albeit using relatively simple benchmark datasets and baseline models. Since the gLMs in these studies were tested upon fine-tuning their weights for each downstream task, determining whether gLM representations embody a foundational understanding of cis-regulatory biology remains an open question. Here we evaluate the representational power of pre-trained gLMs to predict and interpret cell-type-specific functional genomics data that span DNA and RNA regulation. Our findings suggest that current gLMs do not offer substantial advantages over conventional machine learning approaches that use one-hot encoded sequences. This work highlights a major limitation with current gLMs, raising potential issues in conventional pre-training strategies for the non-coding genome.

Improved modeling of RNA-binding protein motifs in an interpretable neural model of RNA splicing

Sequence-specific RNA-binding proteins (RBPs) play central roles in splicing decisions. Here, we describe a modular splicing architecture that leverages in vitro-derived RNA affinity models for 79 human RBPs and the annotated human genome to produce improved models of RBP binding and activity. Binding and activity are modeled by separate Motif and Aggregator components that can be mixed and matched, enforcing sparsity to improve interpretability. Training a new Adjusted Motif (AM) architecture on the splicing task not only yields better splicing predictions but also improves prediction of RBP-binding sites in vivo and of splicing activity, assessed using independent data.

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.

Improved detection of aberrant splicing with FRASER 2.0 and the intron Jaccard index

Detection of aberrantly spliced genes is an important step in RNA-seq-based rare-disease diagnostics. We recently developed FRASER, a denoising autoencoder-based method that outperformed alternative methods of detecting aberrant splicing. However, because FRASER's three splice metrics are partially redundant and tend to be sensitive to sequencing depth, we introduce here a more robust intron-excision metric, the intron Jaccard index, that combines the alternative donor, alternative acceptor, and intron-retention signal into a single value. Moreover, we optimized model parameters and filter cutoffs by using candidate rare-splice-disrupting variants as independent evidence. On 16,213 GTEx samples, our improved algorithm, FRASER 2.0, called typically 10 times fewer splicing outliers while increasing the proportion of candidate rare-splice-disrupting variants by 10-fold and substantially decreasing the effect of sequencing depth on the number of reported outliers. To lower the multiple-testing correction burden, we introduce an option to select the genes to be tested for each sample instead of a transcriptome-wide approach. This option can be particularly useful when prior information, such as candidate variants or genes, is available. Application on 303 rare-disease samples confirmed the relative reduction in the number of outlier calls for a slight loss of sensitivity; FRASER 2.0 recovered 22 out of 26 previously identified pathogenic splicing cases with default cutoffs and 24 when multiple-testing correction was limited to OMIM genes containing rare variants. Altogether, these methodological improvements contribute to more effective RNA-seq-based rare diagnostics by drastically reducing the amount of splicing outlier calls per sample at minimal loss of sensitivity.

Splicing defects in rare diseases: transcriptomics and machine learning strategies towards genetic diagnosis

Genomic variants affecting pre-messenger RNA splicing and its regulation are known to underlie many rare genetic diseases. However, common workflows for genetic diagnosis and clinical variant interpretation frequently overlook splice-altering variants. To better serve patient populations and advance biomedical knowledge, it has become increasingly important to develop and refine approaches for detecting and interpreting pathogenic splicing variants. In this review, we will summarize a few recent developments and challenges in using RNA sequencing technologies for rare disease investigation. Moreover, we will discuss how recent computational splicing prediction tools have emerged as complementary approaches for revealing disease-causing variants underlying splicing defects. We speculate that continuous improvements to sequencing technologies and predictive modeling will not only expand our understanding of splicing regulation but also bring us closer to filling the diagnostic gap for rare disease patients.

Aberrant splicing prediction across human tissues

Aberrant splicing is a major cause of genetic disorders but its direct detection in transcriptomes is limited to clinically accessible tissues such as skin or body fluids. While DNA-based machine learning models can prioritize rare variants for affecting splicing, their performance in predicting tissue-specific aberrant splicing remains unassessed. Here we generated an aberrant splicing benchmark dataset, spanning over 8.8 million rare variants in 49 human tissues from the Genotype-Tissue Expression (GTEx) dataset. At 20% recall, state-of-the-art DNA-based models achieve maximum 12% precision. By mapping and quantifying tissue-specific splice site usage transcriptome-wide and modeling isoform competition, we increased precision by threefold at the same recall. Integrating RNA-sequencing data of clinically accessible tissues into our model, AbSplice, brought precision to 60%. These results, replicated in two independent cohorts, substantially contribute to noncoding loss-of-function variant identification and to genetic diagnostics design and analytics.

The non-coding genome in Autism Spectrum Disorders

Autism Spectrum Disorders (ASD) are a group of neurodevelopmental disorders (NDDs) characterized by difficulties in social interaction and communication, repetitive behavior, and restricted interests. While ASD have been proven to have a strong genetic component, current research largely focuses on coding regions of the genome. However, non-coding DNA, which makes up for ∼99% of the human genome, has recently been recognized as an important contributor to the high heritability of ASD, and novel sequencing technologies have been a milestone in opening up new directions for the study of the gene regulatory networks embedded within the non-coding regions. Here, we summarize current progress on the contribution of non-coding alterations to the pathogenesis of ASD and provide an overview of existing methods allowing for the study of their functional relevance, discussing potential ways of unraveling ASD's "missing heritability".

Improved detection of aberrant splicing using the Intron Jaccard Index

Detection of aberrantly spliced genes is an important step in RNA-seq-based rare disease diagnostics. We recently developed FRASER, a denoising autoencoder-based method for aberrant splicing detection that outperformed alternative approaches. However, as FRASER's three splice metrics are partially redundant and tend to be sensitive to sequencing depth, we introduce here a more robust intron excision metric, the Intron Jaccard Index, that combines alternative donor, alternative acceptor, and intron retention signal into a single value. Moreover, we optimized model parameters and filter cutoffs using candidate rare splice-disrupting variants as independent evidence. On 16,213 GTEx samples, our improved algorithm called typically 10 times fewer splicing outliers while increasing the proportion of candidate rare splice-disrupting variants by 10 fold and substantially decreasing the effect of sequencing depth on the number of reported outliers. Application on 303 rare disease samples confirmed the reduction fold-change of the number of outlier calls for a slight loss of sensitivity (only 2 out of 22 previously identified pathogenic splicing cases not recovered). Altogether, these methodological improvements contribute to more effective RNA-seq-based rare diagnostics by a drastic reduction of the amount of splicing outlier calls per sample at minimal loss of sensitivity.

CPA-Perturb-seq: Multiplexed single-cell characterization of alternative polyadenylation regulators

Most mammalian genes have multiple polyA sites, representing a substantial source of transcript diversity that is governed by the cleavage and polyadenylation (CPA) regulatory machinery. To better understand how these proteins govern polyA site choice we introduce CPA-Perturb-seq, a multiplexed perturbation screen dataset of 42 known CPA regulators with a 3' scRNA-seq readout that enables transcriptome-wide inference of polyA site usage. We develop a statistical framework to specifically identify perturbation-dependent changes in intronic and tandem polyadenylation, and discover modules of co-regulated polyA sites exhibiting distinct functional properties. By training a multi-task deep neural network (APARENT-Perturb) on our dataset, we delineate a cis-regulatory code that predicts responsiveness to perturbation and reveals interactions between distinct regulatory complexes. Finally, we leverage our framework to re-analyze published scRNA-seq datasets, identifying new regulators that affect the relative abundance of alternatively polyadenylated transcripts, and characterizing extensive cellular heterogeneity in 3' UTR length amongst antibody-producing cells. Our work highlights the potential for multiplexed single-cell perturbation screens to further our understanding of post-transcriptional regulation in vitro and in vivo.

Deciphering the impact of genetic variation on human polyadenylation using APARENT2

BACKGROUND

3'-end processing by cleavage and polyadenylation is an important and finely tuned regulatory process during mRNA maturation. Numerous genetic variants are known to cause or contribute to human disorders by disrupting the cis-regulatory code of polyadenylation signals. Yet, due to the complexity of this code, variant interpretation remains challenging.

RESULTS

We introduce a residual neural network model, APARENT2, that can infer 3'-cleavage and polyadenylation from DNA sequence more accurately than any previous model. This model generalizes to the case of alternative polyadenylation (APA) for a variable number of polyadenylation signals. We demonstrate APARENT2's performance on several variant datasets, including functional reporter data and human 3' aQTLs from GTEx. We apply neural network interpretation methods to gain insights into disrupted or protective higher-order features of polyadenylation. We fine-tune APARENT2 on human tissue-resolved transcriptomic data to elucidate tissue-specific variant effects. By combining APARENT2 with models of mRNA stability, we extend aQTL effect size predictions to the entire 3' untranslated region. Finally, we perform in silico saturation mutagenesis of all human polyadenylation signals and compare the predicted effects of [Formula: see text] million variants against gnomAD. While loss-of-function variants were generally selected against, we also find specific clinical conditions linked to gain-of-function mutations. For example, we detect an association between gain-of-function mutations in the 3'-end and autism spectrum disorder. To experimentally validate APARENT2's predictions, we assayed clinically relevant variants in multiple cell lines, including microglia-derived cells.

CONCLUSIONS

A sequence-to-function model based on deep residual learning enables accurate functional interpretation of genetic variants in polyadenylation signals and, when coupled with large human variation databases, elucidates the link between functional 3'-end mutations and human health.

Deep mendelian randomization: Investigating the causal knowledge of genomic deep learning models

Multi-task deep learning (DL) models can accurately predict diverse genomic marks from sequence, but whether these models learn the causal relationships between genomic marks is unknown. Here, we describe Deep Mendelian Randomization (DeepMR), a method for estimating causal relationships between genomic marks learned by genomic DL models. By combining Mendelian randomization with in silico mutagenesis, DeepMR obtains local (locus specific) and global estimates of (an assumed) linear causal relationship between marks. In a simulation designed to test recovery of pairwise causal relations between transcription factors (TFs), DeepMR gives accurate and unbiased estimates of the 'true' global causal effect, but its coverage decays in the presence of sequence-dependent confounding. We then apply DeepMR to examine the global relationships learned by a state-of-the-art DL model, BPNet, between TFs involved in reprogramming. DeepMR's causal effect estimates validate previously hypothesized relationships between TFs and suggest new relationships for future investigation.

Cell-specific regulation of gene expression using splicing-dependent frameshifting

Precise and reliable cell-specific gene delivery remains technically challenging. Here we report a splicing-based approach for controlling gene expression whereby separate translational reading frames are coupled to the inclusion or exclusion of mutated, frameshifting cell-specific alternative exons. Candidate exons are identified by analyzing thousands of publicly available RNA sequencing datasets and filtering by cell specificity, conservation, and local intron length. This method, which we denote splicing-linked expression design (SLED), can be combined in a Boolean manner with existing techniques such as minipromoters and viral capsids. SLED can use strong constitutive promoters, without sacrificing precision, by decoupling the tradeoff between promoter strength and selectivity. AAV-packaged SLED vectors can selectively deliver fluorescent reporters and calcium indicators to various neuronal subtypes in vivo. We also demonstrate gene therapy utility by creating SLED vectors that can target PRPH2 and SF3B1 mutations. The flexibility of SLED technology enables creative avenues for basic and translational research.

Recommendations for clinical interpretation of variants found in non-coding regions of the genome

BACKGROUND

The majority of clinical genetic testing focuses almost exclusively on regions of the genome that directly encode proteins. The important role of variants in non-coding regions in penetrant disease is, however, increasingly being demonstrated, and the use of whole genome sequencing in clinical diagnostic settings is rising across a large range of genetic disorders. Despite this, there is no existing guidance on how current guidelines designed primarily for variants in protein-coding regions should be adapted for variants identified in other genomic contexts.

METHODS

We convened a panel of nine clinical and research scientists with wide-ranging expertise in clinical variant interpretation, with specific experience in variants within non-coding regions. This panel discussed and refined an initial draft of the guidelines which were then extensively tested and reviewed by external groups.

RESULTS

We discuss considerations specifically for variants in non-coding regions of the genome. We outline how to define candidate regulatory elements, highlight examples of mechanisms through which non-coding region variants can lead to penetrant monogenic disease, and outline how existing guidelines can be adapted for the interpretation of these variants.

CONCLUSIONS

These recommendations aim to increase the number and range of non-coding region variants that can be clinically interpreted, which, together with a compatible phenotype, can lead to new diagnoses and catalyse the discovery of novel disease mechanisms.

Genomics enters the deep learning era

The tremendous amount of biological sequence data available, combined with the recent methodological breakthrough in deep learning in domains such as computer vision or natural language processing, is leading today to the transformation of bioinformatics through the emergence of deep genomics, the application of deep learning to genomic sequences. We review here the new applications that the use of deep learning enables in the field, focusing on three aspects: the functional annotation of genomes, the sequence determinants of the genome functions and the possibility to write synthetic genomic sequences.

Predicting RNA splicing from DNA sequence using Pangolin

Recent progress in deep learning has greatly improved the prediction of RNA splicing from DNA sequence. Here, we present Pangolin, a deep learning model to predict splice site strength in multiple tissues. Pangolin outperforms state-of-the-art methods for predicting RNA splicing on a variety of prediction tasks. Pangolin improves prediction of the impact of genetic variants on RNA splicing, including common, rare, and lineage-specific genetic variation. In addition, Pangolin identifies loss-of-function mutations with high accuracy and recall, particularly for mutations that are not missense or nonsense, demonstrating remarkable potential for identifying pathogenic variants.

Splicing in the Diagnosis of Rare Disease: Advances and Challenges

Mutations which affect splicing are significant contributors to rare disease, but are frequently overlooked by diagnostic sequencing pipelines. Greater ascertainment of pathogenic splicing variants will increase diagnostic yields, ending the diagnostic odyssey for patients and families affected by rare disorders, and improving treatment and care strategies. Advances in sequencing technologies, predictive modeling, and understanding of the mechanisms of splicing in recent years pave the way for improved detection and interpretation of splice affecting variants, yet several limitations still prohibit their routine ascertainment in diagnostic testing. This review explores some of these advances in the context of clinical application and discusses challenges to be overcome before these variants are comprehensively and routinely recognized in diagnostics.