Unifying Molecular and Textual Representations via Multi-task Language Modelling

Abstract

The recent advances in neural language models have also been successfully applied to the field of chemistry, offering generative solutions for classical problems in molecular design and synthesis planning. These new methods have the potential to optimize laboratory operations and fuel a new era of data-driven automation in scientific discovery. However, specialized models are still typically required for each task, leading to the need for problem-specific fine-tuning and neglecting task interrelations. The main obstacle in this field is the lack of a unified representation between natural language and chemical representations, complicating and limiting human-machine interaction.Here, we propose a multi-domain, multi-task language model to solve a wide range of tasks in both the chemical and natural language domains. By leveraging multi-task learning, our model can handle chemical and natural language concurrently, without requiring expensive pre-training on single domains or task-specific models. Interestingly, sharing weights across domains remarkably improves our model when benchmarked against state-of-the-art baselines on single-domain and cross-domain tasks. In particular, sharing information across domains and tasks gives rise to large improvements in cross-domain tasks, the magnitude of which increase with scale, as measured by more than a dozen of relevant metrics. Our work suggests that such models can robustly and efficiently accelerate discovery in physical sciences by superseding problem-specific fine-tuning and enhancing human-model interactions.

Instruction Multi-Constraint Molecular Generation Using a Teacher-Student Large Language Model

“…CLAMP [29] introduced a fusion approach, combining a molecule encoder and a text encoder for property prediction tasks. Christofidellis et al [30] presented a unified model capable of handling various text-to-text, text-tomolecule, molecule-to-text, and molecule-to-molecule tasks. MolReGPT [31] implemented tasks such as molecule captioning and text-based molecule generation by assigning ChatGPT a role as a biochemist, facilitating in-context learning.…”

Monte Carlo Thought Search: Large Language Model Querying for Complex Scientific Reasoning in Catalyst Design

“…To enable higher-level control over molecular design, multi-modal models (Edwards et al, 2021;Vall et al, 2021;Zeng et al, 2022;Xu and Wang, 2022;Su et al, 2022;Seidl et al, 2023;Xu et al, 2023;Zhao et al, 2023;Liu et al, 2023b) have been proposed. Existing work focuses on cross-modal retrieval (Edwards et al, 2021;Zeng et al, 2022), translation (Edwards et al, 2022;Liu et al, 2023c;Christofidellis et al, 2023), and editing (Liu et al, 2022).…”