The Goodarzi lab is on fire!

RNA-binding proteins (RBPs) recognize RNA by sequence and shape, but which matters varies by site. Introducing Pythia, an interpretable AI model trained on a new high-resolution atlas of 67 RBPs—a collaboration between Genophoria, Chan Zuckerberg Biohub, UCSF, Calico, & Vector Institute.

What if an AI could look at a strand of RNA, predict how it folds in 3D, and instantly tell you which proteins will attach to it? A groundbreaking new dataset and neural network are doing just that—bringing high-resolution clarity to the invisible mechanics of gene regulation.

Decoding how proteins shape our RNA: a new atlas and AI tool for cellular control.

Inside our cells, DNA serves as the master blueprint, but RNA acts as the working copy that delivers instructions to build proteins. Once RNA is made, it is heavily managed by a special class of molecules called RNA-binding proteins (RBPs). These proteins dictate how long an RNA molecule lasts, where it travels inside the cell, and how efficiently it is translated into functional proteins.

Despite their importance, scientists have lacked high-resolution, uniform maps showing exactly where these proteins bind across the entire human genome.

To bridge this gap, a team of researchers has developed a powerful suite of new experimental and computational tools to decode how proteins and RNA interact.

Key Breakthroughs of the Study.

1. A New Way to Map Protein-RNA Interactions (‘ecCLIP‘):

The researchers created a high-throughput platform called eclair CLIP (ecCLIP). Traditional methods for capturing these interactions often suffer from technical biases and variable efficiency. By combining *endogenous protein tagging* (marking the natural proteins cleanly within human cells) with a streamlined, ligation-free preparation technique, ecCLIP generated crisp, high-resolution binding maps for 67 different RNA-binding proteins in human cells. This created a rich transcriptome-wide atlas that accurately reflects how these proteins regulate RNA stability and lifespan.

2. Improving AI Models of RNA Dynamics (‘Borzuki‘):

To test if this new data could improve predictive biology, the team integrated ecCLIP maps into Borzuki, an advanced sequence-to-function AI model. Incorporating the new binding atlas significantly improved the model’s ability to predict how long specific RNA molecules survive in the cell and how genetic variations in the untranslated regions (3′ UTRs) of RNA can impact stability.

3. Building a Structure-Aware Neural Network (‘Pythia‘):

Proteins don’t just recognize a linear string of RNA letters (sequence); they also care heavily about the three-dimensional shapes that RNA folds into (structure). To model this complex grammar, the researchers designed Pythia, a structure-aware neural network. Pythia uses custom architectural features inspired by context-free grammars to explicitly account for RNA base-pairing. Crucially, the AI can separate predictions into two components: what is driven by the linear sequence versus what is driven by the folded 3D structure.

4. Experimental Validation with Massively Parallel Assays:

To prove that Pythia’s predictions and structural breakdowns were correct, the team used cell-based high-throughput reporter assays (MPRC). By introducing synthetic RNA mutations designed to either disrupt or preserve predicted RNA folding structures, they confirmed that Pythia accurately pinpoints structure-dependent binding sites and successfully predicts how specific genetic variants alter protein-binding behavior.

Implications:

Together, ecCLIP and Pythia provide a powerful combination of experimental data and AI modeling. They offer a clearer, structure-aware window into the hidden rules of gene regulation—helping scientists better understand how post-transcriptional control works and how genetic mutations can disrupt these vital cellular processes.

#ArtificialIntelligence #rnabiology #genomics #Bioinformatics

RNA-binding proteins (RBPs) control post-transcriptional gene expression by regulating RNA processing, localization, stability, and translation. However, our understanding of RNA regulation remains limited by the resolution, scale, and uniformity of available RBP binding maps. Here we introduce eclair CLIP (ecCLIP), a high-throughput CLIP-seq platform that combines endogenous protein tagging for uniform affinity capture with ligation-free library preparation to generate high-resolution RNA-protein binding profiles. We applied ecCLIP to 67 endogenously tagged RBPs in HEK293T cells, generating a high-quality transcriptome-scale atlas that recovered known binding preferences and captured post-transcriptional regulons linked to RNA half-life. To test the utility of the atlas, we incorporated ecCLIP into Borzuki, a Borzoi-derived sequence-to-function model trained on complementary RNA degradation, expression, CLIP, and Perturb-seq data. Inclusion of ecCLIP data improved RNA half-life prediction and assessment of 3′ UTR variant effects on RNA stability, demonstrating that our atlas helps capture RNA dynamics. To directly model the cis-regulatory grammar underlying these RBP-RNA interactions, we developed Pythia, a structure-aware neural network architecture that incorporates context-free grammar-inspired fixed convolutions to explicitly encode RNA base-pairing features. Pythia predicted RBP binding from sequence and structural context and decomposed predictions into sequence-and structure-dependent components. We tested this decomposition for five RBPs using massively parallel reporter CLIP (MPRC) binding assays in cells, using synthetic RNA libraries containing mutations designed to preserve or disrupt predicted RNA structures across nominated binding sites. We used the same MPRC strategy to confirm allele-specific binding effects for variants prioritized by Pythia. Together, ecCLIP and Pythia provide a resource for structure-aware decoding of RNA-protein interaction grammar and interpretation of post-transcriptional regulatory variation.

H.G. acknowledges outside interest as a co-founder of Exai Bio (Network Bio), Tahoe Therapeutics and Therna Biosciences, serves on the board of directors at Exai Bio (Network Bio) and Verge Genomics, and is a Venture Partner at Amplify Partners. D.K. is an employee of Calico Life Sciences LLC. J.L. was previously employed by Calico Life Sciences LLC. B.W. is an employee of Xaira Therapeutics. The remaining authors declare no competing interests.