Accelerating Material Design with the Generative Toolkit for Scientific Discovery

Abstract

With the growing availability of data within various scientific domains, generative models hold enormous potential to accelerate scientific discovery at every step of the scientific method. Perhaps their most valuable application lies in the speeding up of what has traditionally been the slowest and most challenging step of coming up with a hypothesis. Powerful representations are now being learned from large volumes of data to generate novel hypotheses, which is making a big impact on scientific discovery applications ranging from material design to drug discovery. The GT4SD [Team, 2022] (https://github.com/GT4SD/gt4sd-core) is an extensible open-source library that enables scientists, developers and researchers to train and use state-of-the-art generative models for hypothesis generation in scientific discovery. GT4SD supports a variety of uses of generative models across material science and drug discovery, including molecule discovery and design based on properties related to target proteins, omic profiles, scaffold distances, binding energies and more.Keywords Generative Models • Scientific Discovery • Accelerated Discovery • Open Source Humanity's progress has been characterised by a delicate balance between curiosity and creativity. Science is no exception with its long evolution through trial and error. While remarkably successful, the scientific method can be a slow iterative process that can be inadequate when faced with important and pressing needs, e.g., the need to swiftly develop drugs and antibiotics or design novel materials and processes to mitigate climate change effects. Indeed, it can take almost a decade to discover a new material and cost upwards of $10-$100 million. One of the most daunting challenges in materials discovery is hypothesis generation, where it is extremely challenging to identify and select novel and useful candidates in search spaces that are overwhelming in size, e.g., the chemical space for drug-like molecules is estimated to contain 10 33 structures [Polishchuk et al., 2013].To overcome this problem, in recent years, generative models have emerged as an effective approach to design and discover molecules with desired properties. Generative models more efficiently and effectively navigate and explore vast search spaces that are learned from data based on user-defined criteria. Starting from a series of seminal works [Gómez-

On the Choice of Active Site Sequences for Kinase-Ligand Affinity Prediction

“…Moreover, in the Generative Toolkit for Scientific Discovery (GT4SD), we provide an example on leveraging the affinity predictor as a reward function in a protein-driven molecular generative model: .…”

Automated patent classification for crop protection via domain adaptation

“…Additional training epochs did not improve the performance (see Appendix A: Table A7). To train these models, we relied on the Generative Toolkit for Scientific Discovery (GT4SD) library 24 and its LM trainer.…”

Unifying Molecular and Textual Representations via Multi-task Language Modelling

“…We evaluate the model's performance on five tasks: forward and backward reaction prediction in chemistry, textconditional de novo molecule generation and molecule captioning across domains, and paragraph-to-action in the language domain. The training process is carried out using the language modeling trainer based on Hugging Face transformers (Wolf et al, 2020) and PyTorch Lightning (Falcon and The PyTorch Lightning team, 2019) from the GT4SD library (Manica et al, 2022). To initialize our transformer model, we choose to use the natural language domain, as it has the most available data.…”