Generation of antifungals to combat drug resistance using language models and diffusion models.

INTRODUCTION

The growing scarcity of effective antifungal agents, coupled with rising drug‑resistance, creates an urgent demand for novel therapeutics. Recent advances in artificial intelligence (AI) have opened new avenues for accelerating small‑molecule discovery, offering a promising strategy to meet this challenge.

OBJECTIVES

This study aimed to develop an AI-driven pipeline capable of rapidly generating biologically active inhibitors targeting antifungal proteins and to evaluate the therapeutic potential of the resulting candidates.

METHODS

We developed a diffusion-based generative platform, MolDiffusion, to design small molecules against both single- and dual-target profiles. The pipeline was applied to five fungal protein targets and one human protein Keap1. Top-ranked candidate molecules were purchased and evaluated through biochemical assays, cell-based inhibition tests, and in vivo studies using a murine candidiasis model.

RESULTS

Approximately 50% of the candidates generated by the MolDiffusion pipeline exhibited measurable activity in vitro. The platform successfully yielded both single-target and dual-target hits. Notably, two compounds demonstrated significant in vivo efficacy in the mouse model of candidiasis.

CONCLUSION

MolDiffusion effectively translates AI-generated molecular designs into experimentally validated antifungal leads, including dual-target compounds with potential to overcome drug resistance. These findings highlight the platform's promise as a robust tool for next-generation antifungal drug discovery.