arXiv:2503.12602v5 Announce Type: replace 
Abstract: Generative machine learning models for exploring chemical space have shown immense promise, but many molecules they generate are too difficult to synthesize, making them impractical for further investigation or development. In this work, we present a novel approach by fine-tuning Meta's Llama3 Large Language Models (LLMs) to create SynLlama, which generates full synthetic pathways made of commonly accessible building blocks and robust organic reaction templates. SynLlama explores a large synthesizable space using significantly less data, and offers strong performance in both forward and bottom-up synthesis planning compared to other state-of-the-art methods. We find that SynLlama, even without training on external building blocks, can effectively generalize to unseen yet purchasable building blocks, meaning that its reconstruction capabilities extend to a broader synthesizable chemical space than the training data. We also demonstrate the use of SynLlama in a pharmaceutical context for synthesis planning of analog molecules and hit expansion leads for proposed inhibitors of target proteins, offering medicinal chemists a valuable tool for discovery.

SynLlama، الذي طورته شركة ميتا، هو نموذج جديد للتعلم الآلي التوليدي الذي ينشئ جزيئات قابلة للتخليق ونظائرها. من خلال ضبط Llama3، يقوم SynLlama بإنشاء مسارات تخليقية باستخدام كتل بناء شائعة، مما يظهر أداءً قويًا في تخطيط التخليق. هذه الخطوة مهمة لأنها تتيح للكيميائيين الطبيين استكشاف مساحة كيميائية أوسع ببيانات أقل، مما يعزز جهود اكتشاف الأدوية.

SynLlama, desarrollado por Meta, es un nuevo modelo de aprendizaje automático generativo que crea moléculas sintetizables y sus análogos. Al ajustar Llama3, SynLlama genera rutas sintéticas utilizando bloques de construcción comunes, mostrando un sólido rendimiento en la planificación de síntesis. Este avance es significativo ya que permite a los químicos medicinales explorar un espacio químico más amplio con menos datos, mejorando así los esfuerzos de descubrimiento de fármacos.

SynLlama, développé par Meta, est un nouveau modèle d'apprentissage machine génératif qui crée des molécules synthétisables et leurs analogues. En ajustant Llama3, SynLlama génère des voies synthétiques utilisant des blocs de construction courants, montrant de solides performances en planification de synthèse. Cette avancée est significative car elle permet aux chimistes médicaux d'explorer un espace chimique plus large avec moins de données, améliorant ainsi les efforts de découverte de médicaments.

SynLlama, developed by Meta, is a novel generative machine learning model that creates synthesizable molecules and their analogs. By fine-tuning Llama3, SynLlama generates synthetic pathways using common building blocks, demonstrating strong performance in synthesis planning. This advancement is significant as it allows medicinal chemists to explore a broader chemical space with less data, enhancing drug discovery efforts.

SynLlama: Generating Synthesizable Molecules and Their Analogs with Large Language Models

Was this article worth reading? Share it

Ready to build your own newsroom?