Differentiable Scaffolding Tree for Molecular Optimization

By Tianfan Fu, Wenhao Gao, Cao Xiao, Jacob Yasonik, Connor W. Coley, Jimeng Sun. 2021

92 citations recorded by the source corpus.

The structural design of functional molecules, also called molecular optimization, is an essential chemical science and engineering task with important applications, such as drug discovery. Deep generative models and combinatorial optimization methods achieve initial success but still struggle with directly modeling discrete chemical structures and often heavily rely on brute-force enumeration. The challenge comes from the discrete and non-differentiable nature of molecule structures. To address this, we propose differentiable scaffolding tree (DST) that utilizes a learned knowledge network to convert discrete chemical structures to locally differentiable ones. DST enables a gradient-based optimization on a chemical graph structure by back-propagating the derivatives from the target properties through a graph neural network (GNN). Our empirical studies show the gradient-based molecular optimizations are both effective and sample efficient. Furthermore, the learned graph parameters can also provide an explanation that helps domain experts understand the model output.

Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Theory and Mathematics

1. Tianfan Fu
2. Marinka Zitnik
3. Jimeng Sun
4. Shengchao Liu
5. Jure Leskovec
6. Kexin Huang
7. Jian Tang
8. Yuanqi Du

In Materials Chemistry

1. Jure Leskovec
2. Jian Tang
3. Tianfan Fu
4. Jimeng Sun
5. Marinka Zitnik
6. Shengchao Liu
7. Yuanqi Du
8. Yoshua Bengio

In Molecular Biology

1. Petar Velickovic
2. Marinka Zitnik
3. Jure Leskovec
4. Jian Tang
5. Yoshua Bengio
6. Kexin Huang
7. Jimeng Sun
8. Tianfan Fu

Citations in the corpus

• State-specific protein–ligand complex structure prediction with a multiscale deep generative model (2022, positive): “A closed-loop combination of NeuralPLexer and recent advances in differentiable generative models for protein sequences 27,28 , molecular graphs 57,58 and structure-based bioactivity models holds the promise of accelerating the process of sequence and compound prioritization over conventional…”
• Dynamic-Backbone Protein-Ligand Structure Prediction with Multiscale Generative Diffusion Models (2022, background): “…NeuralPLexer only requires molecular graphs as ligand inputs, therefore can enable end-to-end gradient-based design for functional small-molecules and ligand-binding proteins when coupled to recently-proposed differentiable protein sequence [24–26] and molecular graph generators [27, 28].”
• Scientific discovery in the age of artificial intelligence (2023, methodology): “This approach is used in the design of proteins 156 and small molecules 157,158 .”
• MolGenSurvey: A Systematic Survey in Machine Learning Models for Molecule Design (2022, methodology): “There are some early explorations to the interpretability, e.g., [104], [109], [143], [163], [215], [216].”