Roughness of Molecular Property Landscapes and Its Impact on Modellability

Explore the concept of roughness in molecular property landscapes and its impact on modellability in this comprehensive talk by Matteo Aldeghi from Valence Labs. Delve into the introduction of a general, quantitative measure for describing the roughness of molecular property landscapes called the Roughness Index (ROGI). Learn how this index correlates with out-of-sample errors in machine learning models for various regression tasks. Discover the challenges in modeling molecular properties, understand the relationship between roughness and modellability indices, and examine practical examples using 2D datasets. Gain insights into qualitative analysis techniques and methods for predicting machine learning model errors in the context of molecular discovery and drug design.

- Intro
- When is a Molecular Property Challenging to Model?
- Roughness and Modellability Indices
- Measuring Roughness and the Roughness Index ROGI for Molecular Datasets
- ROGI Examples: 2D Datasets
- Qualitative Analysis
- Predicting ML Model Error
- Summary
- Q+A