Researchers at the Massachusetts Institute of Technology have introduced a new computational framework called KATMAP that helps scientists uncover how specific proteins known as splicing factors control the way genes are expressed. Splicing factors play a key role in a process called RNA splicing, where sections of RNA, called exons and introns, are rearranged or removed to produce the final messenger RNA (mRNA) that will be translated into protein. Because different combinations of exons can produce different proteins from the same gene, splicing is one of the main reasons for the complexity of the human genome. When splicing goes wrong, it can lead to serious diseases such as cancer, muscular dystrophy, and neurodegenerative disorders.

RNA sequencing (RNA-seq) experiments typically reveal hundreds of changes in how RNA is spliced, but figuring out why those changes occur is much more difficult. That’s where KATMAP comes in. The software uses an interpretable regression model to connect changes in RNA splicing to changes in splicing factor activity. It takes as input RNA-seq data from experiments where specific splicing factors have been reduced or “knocked down” in cells, along with the known RNA sequence motifs where those factors tend to bind. By analyzing how splicing patterns shift in response to each factor’s knockdown, KATMAP can learn a detailed map of how that factor regulates RNA processing throughout the transcriptome.

High-level overview of the KATMAP model and its applications

a, The general workflow of how KATMAP predicts the regulation of individual exons on the basis of their sequence. b, Learning the rules by which an SF regulates splicing requires only an RNA-seq dataset where that SF was perturbed and some model of its sequence specificity. Bayesian inference determines which splicing models are consistent with the observed splicing changes, yielding an activity map describing the SF’s position-specific regulatory activity and identifying which exons are the SF’s direct targets and which splicing changes represent indirect effects. c, Examples of how learned models can be applied to interpret splicing in terms of SF regulation. Top, scoring an exon of interest to predict the SREs and SFs that control its splicing, with application to variant interpretation and ASO design. Bottom, identifying the causal SFs underlying splicing differences observed in a given RNA-seq dataset.

KATMAP stands out from previous tools because it doesn’t just detect that splicing changes are happening, it identifies which splicing factors are causing them and predicts which exons or RNA regions are directly affected. This makes it especially valuable for distinguishing between direct and indirect regulatory effects, a major challenge in interpreting large-scale RNA-seq datasets.

Another major advantage is that KATMAP comes with pretrained models built from the extensive ENCODE splicing factor knockdown datasets, meaning that researchers can start using it right away without needing to generate their own knockdown data. The software also enables users to interpret splicing changes in clinical RNA-seq samples, helping identify which splicing factors might be active in diseased tissues or cancer cells.

Beyond its analytical capabilities, KATMAP could play an important role in the design of splice-switching antisense oligonucleotides (ASOs), therapeutic molecules that correct harmful splicing events at the RNA level. By predicting which exons are regulated by particular splicing factors, KATMAP can help guide the development of targeted RNA-based therapies.

Overall, KATMAP bridges a critical gap in RNA-seq data analysis by connecting the dots between splicing patterns and the molecular regulators that control them. As our understanding of RNA biology deepens, tools like KATMAP will be essential for translating basic research into new therapeutic strategies.

Availability – The KATMAP library (along with pretrained models) is available from GitLab (https://gitlab.com/LaptopBiologist/katmap).