---
title: "Do Language Models Dream of Binding Molecules? Benchmarking LLMs under Spatial Constraints"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.18144v1"
permalink: "https://feed.astralor.com/items/itm_675453eb1836302121639365"
published: "2026-07-20T16:43:54.000Z"
score: 79
channel: "ai"
tags: ["3D-Fit","LLM","Structure-based Drug Design","Diffusion Models","Spatial Reasoning"]
---

# Do Language Models Dream of Binding Molecules? Benchmarking LLMs under Spatial Constraints

## AI commentary

> This paper is worth attention because it rigorously compares LLMs against diffusion models in 3D molecular generation for the first time. If you are debating between diffusion models and LLMs for drug design, the data and benchmark here provide critical decision support. For AI researchers, it reveals LLM weaknesses and potential in spatial reasoning, inspiring new architectures.

## At a glance

This paper systematically analyzes the ability of general-purpose large language models (LLMs) to handle 3D spatial constraints in structure-based drug design (SBDD). Compared to specialized models like diffusion models, LLMs still lag behind in multi-conditioned spatial molecule generation tasks such as protein pocket constraints, pharmacophore points, and mandatory interactions, but show promise in handling multiple spatial constraints simultaneously. The authors introduce 3D-Fit, a token-efficient benchmarking strategy, to evaluate LLMs and reveal their scalability to heterogeneous setups.

## Key points

- LLMs lag behind diffusion models in multi-conditioned spatial molecule generation but show potential to handle multiple constraints simultaneously.
- Introduce 3D-Fit benchmark for efficient evaluation of LLMs under complex spatial constraints.
- Findings reveal LLM spatial reasoning capabilities are scalable to heterogeneous setups, laying groundwork for unified models.
- Systematic comparison between general-purpose LLMs and specialized diffusion models in drug design.

## Full text

Structure-based drug design (SBDD) leverages the 3D structure of protein targets, often complemented by other spatial constraints, to generate candidate binding molecules. While diffusion models have dominated as a leading paradigm for high-quality 3D molecule generation, LLM-based methods are rapidly emerging in molecular design and have shown competitive performance in pocket-conditioned molecular generation. However, their ability to reason about physics and 3D spatial environments is largely underexplored. In this work, we systematically analyze whether current general-purpose LLMs are capable of navigating complex 3D constraints compared to established baselines such as specialized diffusion models. We consider 3D ligand generation conditioned on protein pockets together with ligand- and interaction-derived spatial constraints, including anchor fragments, pharmacophore points, and mandatory pocket-ligand interactions. To enable this evaluation, we introduce 3D-Fit - a token-efficient benchmarking strategy for assessing LLM performance on multi-conditioned spatial molecule generation. Our findings reveal a clear pattern in LLM spatial capabilities: while they still lag behind state-of-the-art approaches, they are promising and can handle multiple spatial constraints simultaneously, enabling scaling to heterogeneous setups.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_675453eb1836302121639365