arXiv:2603.12344v2 Announce Type: replace 
Abstract: Molecular Property Prediction (MPP) is a fundamental problem in drug discovery that has recently attracted growing attention. Large Language Models (LLMs), known for their impressive proficiency across domains, show promise as generalist models for MPP. However, their current performance remains below the threshold needed for practical adoption. To bridge this gap, we propose TreeKD for distilling the knowledge of tree-based specialist models into LLMs to complement the internal knowledge of LLMs and improve their predictive accuracy. For each property, we train a specialist decision tree using features derived from 40K functional groups in the input molecules. Then, the predictive rule learned by the decision tree, which encodes its knowledge, is verbalized and incorporated into the prompts for training LLMs. In addition, by replacing a single decision tree with a Random Forest, we introduce a test-time scaling technique called rule-consistency, which aggregates predictions generated from different prompts constructed with different rules. An extensive evaluation with two LLMs, Gemma-2-2B and Granite-3.3-2B, on the TDC benchmark with 22 prediction tasks shows that our method substantially enhances the performance of LLMs, advancing the development of generalist models for MPP.

اقترحت أبحاث حديثة نهجًا جديدًا يسمى TreeKD، والذي يهدف إلى تحسين دقة التنبؤ لنماذج اللغة الكبيرة (LLMs) في التنبؤ بخصائص الجزيئات (MPP) من خلال تقطير المعرفة من أشجار القرار. تتضمن هذه الطريقة تدريب أشجار قرار متخصصة على ميزات مستمدة من 40,000 مجموعة وظيفية في الجزيئات وتدوين قواعدها التنبؤية لتدريب LLMs.

Investigaciones recientes han propuesto un enfoque novedoso llamado TreeKD, que busca mejorar la precisión predictiva de los Modelos de Lenguaje de Gran Tamaño (LLMs) en la Predicción de Propiedades Moleculares (MPP) mediante la destilación de conocimientos de árboles de decisión. Este método implica entrenar árboles de decisión especializados con características derivadas de 40,000 grupos funcionales en las moléculas y verbalizar sus reglas predictivas para el entrenamiento de los LLMs.

Des recherches récentes ont proposé une nouvelle approche appelée TreeKD, qui vise à améliorer la précision prédictive des modèles de langage de grande taille (LLMs) dans la prédiction des propriétés moléculaires (MPP) en distillant les connaissances des arbres de décision. Cette méthode consiste à former des arbres de décision spécialisés sur des caractéristiques dérivées de 40 000 groupes fonctionnels dans les molécules et à verbaliser leurs règles prédictives pour l'entraînement des LLMs.

Recent research has proposed a novel approach called TreeKD, which aims to enhance the predictive accuracy of Large Language Models (LLMs) in Molecular Property Prediction (MPP) by distilling knowledge from decision trees. This method involves training specialist decision trees on features derived from 40,000 functional groups in molecules and verbalizing their predictive rules for LLM training.