MolRAG: Unlocking the Power of Large Language Models for Molecular Property Prediction

Recent LLMs exhibit limited effectiveness on molecular property prediction task due to the semantic gap between molecular representations and natural language, as well as the lack of domain-specific knowledge.To address these challenges, we propose MolRAG, a Retrieval-Augmented Generation framework integrating Chain-of-Thought reasoning for molecular property prediction.MolRAG operates by retrieving structurally analogous molecules as contextual references to guide stepwise knowledge reasoning through chemical structureproperty relationships.This dual mechanism synergizes molecular similarity analysis with structured inference, while generating humaninterpretable rationales grounded in domain knowledge.Experimental results show Mol-RAG outperforms pre-trained LLMs on four datasets, and even matches supervised methods, achieving performance gains of 1.1%-45.7%over direct prediction approaches, demonstrating versatile effectiveness.Our code is available at https://github.com/AcaciaSin/MolRAG.* Molecule 1: No 3D structure, no sulfone group, no aromatic ring * Molecule 2: Sulfone group, aromatic ring, but no 3D structure * Molecule 3: Sulfone group, aromatic ring, 3D structure, but no nitrogen-containing ring * Molecule 4: No sulfone group, aromatic ring, no 3D structure Now, let's analyze the question input molecule: O=SThis molecule has: * A sulfone group (S(=O)(=O)O) * An aromatic ring (c1cc2...) * A 3D structure (with a nitrogen-containing ring) * A sulfone group interacting with the nitrogen-containing ring Comparing the question input molecule to the examples, I notice that it has a sulfone group, an aromatic ring, and a 3D structure, which are similar to Molecule 3.However, the question input molecule has a nitrogen-containing ring, which is not present in Molecule 3. * Has fluorine atoms (C(F)(F)F) which may affect its properties.* Score: 0.6575, Label: No Example 2: * Has a different structure, with a carbonyl group (C(=O)CC(c2ccccc2)) and a sulfonamide group (Sc2ccccc21).* Score: 0.5634, Label: No Example 3: * Has a different structure, with a carbonyl group (C(=O)c2cc3ccccc3cc21) and no sulfonamide group.* Score: 0.5397, Label: No Example 4: * Has a different structure, with a hydroxyl group (O) and a bromine atom (Br), but no sulfonamide group.* Score: 0.5312, Label: No Now, let's analyze the input molecule: * It has a sulfonamide group (Sc2c1ccc1ccccc21) and a carbonyl group (CC(=O)N1c2ccccc2).* It has no fluorine atoms, unlike Example 1. * It has a relatively simple structure with a single aromatic ring and a sulfonamide group.

Authors

Institutions

Publication Details

Published
2025-01-01
DOI
https://doi.org/10.18653/v1/2025.acl-long.755
Citations
4
Primary Topic
Computational Drug Discovery Methods
Type
article
Field-Weighted Citation Impact
6.02
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

MolRAG: Unlocking the Power of Large Language Models for Molecular Property Prediction

Shangsong Liang, Ziting Xian, Jing Gu, Lingbo Li
4 citations
Computational Drug Discovery Methods
6.02
article

MolRAG: Unlocking the Power of Large Language Models for Molecular Property Prediction

Shangsong Liang, Ziting Xian, Jing Gu, Lingbo Li
article en
4 citations

Abstract

Recent LLMs exhibit limited effectiveness on molecular property prediction task due to the semantic gap between molecular representations and natural language, as well as the lack of domain-specific knowledge.To address these challenges, we propose MolRAG, a Retrieval-Augmented Generation framework integrating Chain-of-Thought reasoning for molecular property prediction.MolRAG operates by retrieving structurally analogous molecules as contextual references to guide stepwise knowledge reasoning through chemical structureproperty relationships.This dual mechanism synergizes molecular similarity analysis with structured inference, while generating humaninterpretable rationales grounded in domain knowledge.Experimental results show Mol-RAG outperforms pre-trained LLMs on four datasets, and even matches supervised methods, achieving performance gains of 1.1%-45.7%over direct prediction approaches, demonstrating versatile effectiveness.Our code is available at https://github.com/AcaciaSin/MolRAG.* Molecule 1: No 3D structure, no sulfone group, no aromatic ring * Molecule 2: Sulfone group, aromatic ring, but no 3D structure * Molecule 3: Sulfone group, aromatic ring, 3D structure, but no nitrogen-containing ring * Molecule 4: No sulfone group, aromatic ring, no 3D structure Now, let's analyze the question input molecule: O=SThis molecule has: * A sulfone group (S(=O)(=O)O) * An aromatic ring (c1cc2...) * A 3D structure (with a nitrogen-containing ring) * A sulfone group interacting with the nitrogen-containing ring Comparing the question input molecule to the examples, I notice that it has a sulfone group, an aromatic ring, and a 3D structure, which are similar to Molecule 3.However, the question input molecule has a nitrogen-containing ring, which is not present in Molecule 3. * Has fluorine atoms (C(F)(F)F) which may affect its properties.* Score: 0.6575, Label: No Example 2: * Has a different structure, with a carbonyl group (C(=O)CC(c2ccccc2)) and a sulfonamide group (Sc2ccccc21).* Score: 0.5634, Label: No Example 3: * Has a different structure, with a carbonyl group (C(=O)c2cc3ccccc3cc21) and no sulfonamide group.* Score: 0.5397, Label: No Example 4: * Has a different structure, with a hydroxyl group (O) and a bromine atom (Br), but no sulfonamide group.* Score: 0.5312, Label: No Now, let's analyze the input molecule: * It has a sulfonamide group (Sc2c1ccc1ccccc21) and a carbonyl group (CC(=O)N1c2ccccc2).* It has no fluorine atoms, unlike Example 1. * It has a relatively simple structure with a single aromatic ring and a sulfonamide group.

National Sun Yat-sen University (TW), Sun Yat-sen University (CN), Sun Yat-sen Memorial Hospital (CN)
Openalex Percentile: Top 4%
Computational Drug Discovery Methods
6.02
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.