Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rethinking Specificity in SBDD: Leveraging Delta Score and Energy-Guided Diffusion

Bowen Gao, Minsi Ren, Yuyan Ni, Yanwen Huang, Bo Qiang, Zhi-Ming Ma, Wei-Ying Ma, Yanyan Lan

TL;DR

An innovative energy-guided approach using contrastive learning, with active compounds as decoys, to direct generative models toward creating molecules with high specificity is developed, successfully bridging the gap between SBDD and real-world needs.

Abstract

In the field of Structure-based Drug Design (SBDD), deep learning-based generative models have achieved outstanding performance in terms of docking score. However, further study shows that the existing molecular generative methods and docking scores both have lacked consideration in terms of specificity, which means that generated molecules bind to almost every protein pocket with high affinity. To address this, we introduce the Delta Score, a new metric for evaluating the specificity of molecular binding. To further incorporate this insight for generation, we develop an innovative energy-guided approach using contrastive learning, with active compounds as decoys, to direct generative models toward creating molecules with high specificity. Our empirical results show that this method not only enhances the delta score but also maintains or improves traditional docking scores, successfully bridging the gap between SBDD and real-world needs.

Rethinking Specificity in SBDD: Leveraging Delta Score and Energy-Guided Diffusion

TL;DR

An innovative energy-guided approach using contrastive learning, with active compounds as decoys, to direct generative models toward creating molecules with high specificity is developed, successfully bridging the gap between SBDD and real-world needs.

Abstract

In the field of Structure-based Drug Design (SBDD), deep learning-based generative models have achieved outstanding performance in terms of docking score. However, further study shows that the existing molecular generative methods and docking scores both have lacked consideration in terms of specificity, which means that generated molecules bind to almost every protein pocket with high affinity. To address this, we introduce the Delta Score, a new metric for evaluating the specificity of molecular binding. To further incorporate this insight for generation, we develop an innovative energy-guided approach using contrastive learning, with active compounds as decoys, to direct generative models toward creating molecules with high specificity. Our empirical results show that this method not only enhances the delta score but also maintains or improves traditional docking scores, successfully bridging the gap between SBDD and real-world needs.
Paper Structure (29 sections, 1 theorem, 26 equations, 6 figures, 3 tables, 1 algorithm)

This paper contains 29 sections, 1 theorem, 26 equations, 6 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Reassessing SBDD: Addressing Molecular Specificity
  4. Notable Findings in Docking Scores
  5. Specificity Concerns in SBDD
  6. Inadequacy of Docking Scores to Reflect Specificity
  7. Limitations in Modeling Specificity in Previous SBDD Methods
  8. Method
  9. From Unconditional to Conditional
  10. Measure the specific binding
  11. Generation with Specific binding
  12. Overview of our method
  13. Molecular Diffusion Model
  14. Specific Binding Energy Function
  15. SB-Energy Minimizer Diffusion Model
  16. ...and 14 more sections

Key Result

Lemma 1.1

For a finite set of real numbers $z_1, z_2, \cdots, z_N$, the log-sum-exp function can be upper and lower bounded by the maximum function.

Figures (6)

  • Figure 1: Docking Scores against targets in the test set after sorted (multiplied by -1)
  • Figure 2: The number of pockets which surpass true target on docking score
  • Figure 4: The main framework of our SBE-Diff model. Part a) depicts the training phase of the Specific Binding Energy (SBE) model, where in-batch softmax loss is employed. We use ligands designated for different pockets as negative instances for the current pocket, aiming to address the specific binding. Part b) describes the model's reverse diffusion process, wherein the transition from $X_{t+1}$ to $X_t$ is guided by minimizing the specific binding energy. Part c) demonstrates the calculation of the SBE. This is achieved by utilizing pocket and molecule embeddings, produced by their respective specially trained encoders.
  • Figure 5: The evolution of absolute docking scores and delta scores obtained by various methods, organized chronologically.
  • Figure 6: Case study of generated molecules docked with its original target and shuffled target.
  • ...and 1 more figures

Theorems & Definitions (2)

  • Lemma 1.1: Log-sum-exp approximation
  • proof