Genes contain the instructions for building proteins, but those instructions are not always used in the same way. Through a process called alternative splicing, cells can produce multiple RNA transcripts and protein variants from a single gene. This mechanism helps create the diversity needed for different tissues and cell types to perform specialized functions.
Because alternative splicing plays a major role in normal biology and many diseases, scientists are increasingly interested in predicting and controlling splicing outcomes. Researchers from the University of Pennsylvania have developed a new artificial intelligence framework called TrASPr+BOS that can both predict and design tissue-specific RNA splicing patterns.
RNA alternative splicing (AS) and its predictive generative modeling
(a) Cartoons of several types of AS, involving exon skipping as well as alternative 3’ and 5’ splice sites. (b) Quantification of AS events such as exon skipping from RNA-Seq. PSI is used to represent the relative exon inclusion level, and dPSI is used to show the inclusion change across different conditions. Quantifying thousands of such AS events across many conditions, represented by the 3D stacks, enables training ML algorithms to predict PSI and dPSI from genomic sequence. (c) A genome browser view of an illustrative exon skipping event. The genomic regions spanned by cassette exons vary from tens to hundreds of thousands of bases. (d) Schematic of TrASPr (left) and BOS (right). TrASPr’s input includes genomic regions proximal to splice sites around AS event e, along with other genomic features and two cellular conditions (c,c′; bottom left). It outputs a predicted inclusion level Ψ^e,c as well as predictions for increased inclusion and increased exclusion between the two cellular conditions (ΔΨ^+e,c,c′,ΔΨ^−e,c,c′). Its predictions are then used to train the generative model using BOS. See main text for details.
The system combines two complementary technologies. The first component, called Transformer for Alternative Splicing Prediction, or TrASPr, uses transformer-based artificial intelligence models to predict how RNA sequences will be spliced in different tissues and biological conditions.
Unlike many existing tools, TrASPr can analyze multiple types of alternative splicing events and can generalize to cellular conditions that were not included during training. This makes the model more flexible for studying complex biological systems.
The second component, called Bayesian Optimization for Splicing, or BOS, takes the predictions generated by TrASPr and uses them to design RNA sequences that produce specific splicing outcomes. In other words, the framework does not simply predict what will happen, it can also suggest sequence modifications that may achieve a desired result.
The researchers found that TrASPr+BOS outperformed existing approaches for tissue-specific splicing prediction. In some cases, performance improved by as much as 1.8-fold when measuring tissue-specific prediction accuracy.
An important aspect of the work involved identifying regulatory elements that influence alternative splicing. The model successfully detected tissue-specific sequence features that help determine how RNA molecules are processed in different biological contexts.
To validate the predictions, the researchers examined hundreds of newly predicted tissue-specific splicing variations. They also experimentally confirmed several previously unknown regulatory elements using dCas13-based approaches, providing evidence that the model was identifying biologically meaningful signals.
The ability to design RNA sequences with specific splicing behaviors could have significant implications for biotechnology and medicine. Many genetic disorders are linked to abnormal splicing patterns, and future therapies may seek to correct these defects by altering RNA processing.
Artificial intelligence tools such as TrASPr+BOS may also help researchers better understand gene regulation, discover new therapeutic targets, and accelerate the development of RNA-based medicines.
As RNA biology continues to become more important in modern medicine, predictive and generative AI models are emerging as valuable tools for exploring how genes are regulated across different tissues and disease states.
Availability – Source code and data is available on https://bitbucket.org/biociphers/traspr/src/main/
Wu D, Maus N, Jha A, Yang K, Wales-McGrath BD, Jewell S, Tangiyan A, Choi P, Gardner JR, Barash Y. (2026) Generative modeling for RNA splicing prediction and design. eLife 14:RP106043. [article]
Related Posts
Genes contain the instructions for building proteins, but those instructions are not always used in the same way. Through a process called alternative splicing, cells can produce multiple RNA transcripts and protein variants from a single gene. This mechanism helps create the diversity needed for different tissues and cell types to perform specialized functions.
Because alternative splicing plays a major role in normal biology and many diseases, scientists are increasingly interested in predicting and controlling splicing outcomes. Researchers from the University of Pennsylvania have developed a new artificial intelligence framework called TrASPr+BOS that can both predict and design tissue-specific RNA splicing patterns.
RNA alternative splicing (AS) and its predictive generative modeling
(a) Cartoons of several types of AS, involving exon skipping as well as alternative 3’ and 5’ splice sites. (b) Quantification of AS events such as exon skipping from RNA-Seq. PSI is used to represent the relative exon inclusion level, and dPSI is used to show the inclusion change across different conditions. Quantifying thousands of such AS events across many conditions, represented by the 3D stacks, enables training ML algorithms to predict PSI and dPSI from genomic sequence. (c) A genome browser view of an illustrative exon skipping event. The genomic regions spanned by cassette exons vary from tens to hundreds of thousands of bases. (d) Schematic of TrASPr (left) and BOS (right). TrASPr’s input includes genomic regions proximal to splice sites around AS event e, along with other genomic features and two cellular conditions (c,c′; bottom left). It outputs a predicted inclusion level Ψ^e,c as well as predictions for increased inclusion and increased exclusion between the two cellular conditions (ΔΨ^+e,c,c′,ΔΨ^−e,c,c′). Its predictions are then used to train the generative model using BOS. See main text for details.
The system combines two complementary technologies. The first component, called Transformer for Alternative Splicing Prediction, or TrASPr, uses transformer-based artificial intelligence models to predict how RNA sequences will be spliced in different tissues and biological conditions.
Unlike many existing tools, TrASPr can analyze multiple types of alternative splicing events and can generalize to cellular conditions that were not included during training. This makes the model more flexible for studying complex biological systems.
The second component, called Bayesian Optimization for Splicing, or BOS, takes the predictions generated by TrASPr and uses them to design RNA sequences that produce specific splicing outcomes. In other words, the framework does not simply predict what will happen, it can also suggest sequence modifications that may achieve a desired result.
The researchers found that TrASPr+BOS outperformed existing approaches for tissue-specific splicing prediction. In some cases, performance improved by as much as 1.8-fold when measuring tissue-specific prediction accuracy.
An important aspect of the work involved identifying regulatory elements that influence alternative splicing. The model successfully detected tissue-specific sequence features that help determine how RNA molecules are processed in different biological contexts.
To validate the predictions, the researchers examined hundreds of newly predicted tissue-specific splicing variations. They also experimentally confirmed several previously unknown regulatory elements using dCas13-based approaches, providing evidence that the model was identifying biologically meaningful signals.
The ability to design RNA sequences with specific splicing behaviors could have significant implications for biotechnology and medicine. Many genetic disorders are linked to abnormal splicing patterns, and future therapies may seek to correct these defects by altering RNA processing.
Artificial intelligence tools such as TrASPr+BOS may also help researchers better understand gene regulation, discover new therapeutic targets, and accelerate the development of RNA-based medicines.
As RNA biology continues to become more important in modern medicine, predictive and generative AI models are emerging as valuable tools for exploring how genes are regulated across different tissues and disease states.
Availability – Source code and data is available on https://bitbucket.org/biociphers/traspr/src/main/
Wu D, Maus N, Jha A, Yang K, Wales-McGrath BD, Jewell S, Tangiyan A, Choi P, Gardner JR, Barash Y. (2026) Generative modeling for RNA splicing prediction and design. eLife 14:RP106043. [article]
Related Posts
Recent Posts
Novel biosensor uses DNA ‘barcodes’ to uncover disease signals in cells
GP-DHT uses single-cell RNA sequencing to map gene regulatory networks
Boehringer taps Envisagenics’ AI to mine RNA splicing for cancer targets
CellART extracts single-cell information from high-resolution spatial transcriptomics
Cell2space restores spatial context to single-cell RNA sequencing data
PacBio and TychoBio Collaborate to Train AI Models for Rare Disease
RNA-Seq Products & Services
Stay Connected