Chemical Language Models for Natural Products: A State-Space Model Approach

TL;DR

This paper develops and compares state-space and transformer-based chemical language models tailored for natural products, demonstrating that domain-specific pre-training on a modest dataset can achieve competitive performance in molecule generation and property prediction.

Contribution

It introduces NP-specific state-space models (Mamba and Mamba-2), systematically compares them with transformers, and evaluates various tokenization strategies for natural product tasks.

Findings

Mamba generates 1-2% more valid and unique molecules than GPT.

Mamba variants outperform GPT in property prediction by 0.02-0.04 MCC.

Domain-specific pre-training on 1M NPs matches larger dataset models.

Abstract

Language models are widely used in chemistry for molecular property prediction and small-molecule generation, yet Natural Products (NPs) remain underexplored despite their importance in drug discovery. To address this gap, we develop NP-specific chemical language models (NPCLMs) by pre-training state-space models (Mamba and Mamba-2) and comparing them with transformer baselines (GPT). Using a dataset of about 1M NPs, we present the first systematic comparison of selective state-space models and transformers for NP-focused tasks, together with eight tokenization strategies including character-level, Atom-in-SMILES (AIS), byte-pair encoding (BPE), and NP-specific BPE. We evaluate molecule generation (validity, uniqueness, novelty) and property prediction (membrane permeability, taste, anti-cancer activity) using MCC and AUC-ROC. Mamba generates 1-2 percent more valid and unique molecules than Mamba-2 and GPT, with fewer long-range dependency errors, while GPT yields slightly more novel structures. For property prediction, Mamba variants outperform GPT by 0.02-0.04 MCC under random splits, while scaffold splits show comparable performance. Results demonstrate that domain-specific pre-training on about 1M NPs can match models trained on datasets over 100 times larger.