Alexander Schug: Deciphering RNA: From Statistical Signals to Foundation Models

Speaker:
Alexander Schug (KIT Karlsruhe)

Title:
Deciphering RNA: From Statistical Signals to Foundation Models

Abstract

Proteins and RNA are fundamental to life, and predicting their structure and function from sequence remains a central challenge. For proteins, statistical signals of co-evolution opened the door to accurate contact prediction, culminating in AlphaFold and its revolution for structural biology. RNA presents an even greater frontier: while it underpins regulation, catalysis, and disease, experimental structures are scarce, making a direct transfer of protein methods not possible. In this talk, I will chart the trajectory from statistical-physics based models to deep learning AI approaches, highlighting how principles of molecular interactions can be captured from sequence data alone. I will introduce NucleicBERT, a large-scale, alignment-free language model trained on 30 million RNA sequences. NucleicBERT achieves state-of-the-art predictions for structure and function and, intriguingly, organizes RNA families and functional motifs without supervision thereby effectively “rediscovering” RNA biology. Looking forward, such models offer a powerful way to uncover the physical and functional rules encoded in RNA sequences, promising to bridge abundant genomic data with the mechanistic understanding needed for biomedicine.

Read more about his research here.

Lecture Series:
SysBio Lecture Series: AI for Systems Medicine