Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Conformation Generation using Transformer Flows

Sohil Atul Shah, Vladlen Koltun

TL;DR

This work tackles fast and accurate generation of valid 3D conformations for molecular graphs. It introduces ConfFlow, a graph-conditioned continuous normalizing flow that directly samples 3D coordinates from a simple base distribution, without enforcing explicit geometric constraints. The method combines graph-conditioned CNFs with a Transformer-based Graph Continuous Flow (GCF) architecture, stabilized by kinetic and Jacobian regularization, and trained to maximize exact likelihoods. Empirical results on GEOM-QM9 and GEOM-Drugs show ConfFlow achieving up to 40% improvements over state-of-the-art baselines in conformation generation and competitive performance on ensemble property predictions, highlighting its scalability to large drug-like molecules. The work provides a practical, non-equivariant coordinate-space generative approach with code availability for community use.</n>

Abstract

Estimating three-dimensional conformations of a molecular graph allows insight into the molecule's biological and chemical functions. Fast generation of valid conformations is thus central to molecular modeling. Recent advances in graph-based deep networks have accelerated conformation generation from hours to seconds. However, current network architectures do not scale well to large molecules. Here we present ConfFlow, a flow-based model for conformation generation based on transformer networks. In contrast with existing approaches, ConfFlow directly samples in the coordinate space without enforcing any explicit physical constraints. The generative procedure is highly interpretable and is akin to force field updates in molecular dynamics simulation. When applied to the generation of large molecule conformations, ConfFlow improve accuracy by up to $40\%$ relative to state-of-the-art learning-based methods. The source code is made available at https://github.com/IntelLabs/ConfFlow.

Conformation Generation using Transformer Flows

TL;DR

This work tackles fast and accurate generation of valid 3D conformations for molecular graphs. It introduces ConfFlow, a graph-conditioned continuous normalizing flow that directly samples 3D coordinates from a simple base distribution, without enforcing explicit geometric constraints. The method combines graph-conditioned CNFs with a Transformer-based Graph Continuous Flow (GCF) architecture, stabilized by kinetic and Jacobian regularization, and trained to maximize exact likelihoods. Empirical results on GEOM-QM9 and GEOM-Drugs show ConfFlow achieving up to 40% improvements over state-of-the-art baselines in conformation generation and competitive performance on ensemble property predictions, highlighting its scalability to large drug-like molecules. The work provides a practical, non-equivariant coordinate-space generative approach with code availability for community use.</n>

Abstract

Estimating three-dimensional conformations of a molecular graph allows insight into the molecule's biological and chemical functions. Fast generation of valid conformations is thus central to molecular modeling. Recent advances in graph-based deep networks have accelerated conformation generation from hours to seconds. However, current network architectures do not scale well to large molecules. Here we present ConfFlow, a flow-based model for conformation generation based on transformer networks. In contrast with existing approaches, ConfFlow directly samples in the coordinate space without enforcing any explicit physical constraints. The generative procedure is highly interpretable and is akin to force field updates in molecular dynamics simulation. When applied to the generation of large molecule conformations, ConfFlow improve accuracy by up to relative to state-of-the-art learning-based methods. The source code is made available at https://github.com/IntelLabs/ConfFlow.

Paper Structure

This paper contains 19 sections, 11 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Formulation
  3. Graph-conditioned Normalizing Flows
  4. GCF: Graph Continuous Flows
  5. Regularization
  6. Training Objective
  7. Sampling
  8. Implementation
  9. Experiments
  10. Data
  11. Baselines
  12. Conformation Generation
  13. Property Prediction
  14. Conclusion
  15. Additional Experiments
  16. ...and 4 more sections

Figures (3)

  • Figure 1: ConfFlow transforms points sampled from a simple prior to a 3D molecular conformation. Videos are provided in the supplement.
  • Figure 2: Illustration of ConfFlow with an architectural overview (left), a graph continuous flow (GCF) block (middle), and a graph conditional point transformer (GCPT) message passing layer used within GCF (right).
  • Figure 3: Visualization of conformations generated by CGCF, ConfGF, and ConfFlow (our approach) for four randomly selected molecular graphs from the test set of GEOM-Drugs. Following a similar protocol to Table \ref{['tab:confgen']}, for each approach we sample $2\times$ conformations and pick the one that aligns best with the reference.