Explore a comprehensive talk on designing DNA with tunable regulatory activity using discrete diffusion. Delve into the innovative DNA Discrete Diffusion (D3) framework for conditionally sampling regulatory sequences with targeted functional activity levels. Learn about the challenges in engineering regulatory DNA sequences and how D3 outperforms existing methods in capturing cis-regulatory grammars. Discover a suite of evaluation metrics assessing functional similarity, sequence similarity, and regulatory composition of generated sequences. Understand how D3-generated sequences can augment supervised models and improve predictive performance in data-limited scenarios. Follow the speakers through key topics including continuous time framework, concrete score, learning process, sampling process, and evaluation methods. Engage with the Q&A session to gain deeper insights into this cutting-edge research in AI for drug discovery.
Overview
Syllabus
- Intro + Background
- Continuous Time Framework
- Concrete Score
- Learning Concrete Scores
- Training Process
- Sampling Process
- Evaluation
- Conclusions
- Q+A
Taught by
Valence Labs
