Pesti-Gen: Unleashing a Generative Molecule Approach for Toxicity Aware Pesticide Design

TL;DR

Pesti-Gen tackles the problem of generating pesticide candidates with toxicity-aware constraints, addressing the gap where prior ML focused on prediction rather than generation. It employs a two-stage approach: a latent-space learning phase using a variational autoencoder on general SMILES, followed by toxicity-guided fine-tuning that optimizes for livestock ($LD_{50}$) and aquatic ($LC_{50}$) toxicity via a multi-objective loss $L' = L_{KL} + L_{recon} + \alpha L_{toxicity}^{(livestock)} + \beta L_{toxicity}^{(aqua)}$ with $\alpha=\beta=0.5$. The authors assemble a sizable, task-specific dataset (56,360 data points, 520 pesticides) from LD50/LC50 data and the Gyeonggi Province Open Dataset to enable robust toxicity conditioning. Results show the model achieves about 68% SMILES validity, produces chemically diverse candidates with favorable LogP and SAS profiles, and preserves core scaffolds while incorporating eco-friendly modifications, showcasing a practical path toward safer pesticide design. Limitations include data licensing constraints and the need for wet-lab validation, but the work lays a scalable foundation for eco-conscious, multi-objective pesticide discovery.

Abstract

Global climate change has reduced crop resilience and pesticide efficacy, making reliance on synthetic pesticides inevitable, even though their widespread use poses significant health and environmental risks. While these pesticides remain a key tool in pest management, previous machine-learning applications in pesticide and agriculture have focused on classification or regression, leaving the fundamental challenge of generating new molecular structures or designing novel candidates unaddressed. In this paper, we propose Pesti-Gen, a novel generative model based on variational auto-encoders, designed to create pesticide candidates with optimized properties for the first time. Specifically, Pesti-Gen leverages a two-stage learning process: an initial pre-training phase that captures a generalized chemical structure representation, followed by a fine-tuning stage that incorporates toxicity-specific information. The model simultaneously optimizes over multiple toxicity metrics, such as (1) livestock toxicity and (2) aqua toxicity to generate environmentally friendly pesticide candidates. Notably, Pesti-Gen achieves approximately 68\% structural validity in generating new molecular structures, demonstrating the model's effectiveness in producing optimized and feasible pesticide candidates, thereby providing a new way for safer and more sustainable pest management solutions.

Figures (2)

• Figure 1: Overview of the two-stage Pesti-Gen training process: (1) Latent Space Training learns a general molecular representation using reconstruction and KL loss; (2) Fine-Tuning on Properties incorporates toxicity-specific properties ($\textit{LD}_{50}$ and ecological toxicity) with targeted loss components to optimize pesticide candidates.
• Figure 2: Samples of reconstructed pesticide molecules generated by the trained model. Each pair consists of an original molecule (left) and its corresponding reconstructed molecule (right).