Cheminformatics

Molexar: A Unified Multimodal Molecular Foundation Model for Drug Design

Haoyu Lin, Yiyan Liao, Jinmei Pan, Xinliao Ling, Luhua Lai, Jianfeng Pei

Featured July 9, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

A new AI model called Molexar uses a special molecular language and a clever trick to inject different design rules directly into its input, letting it efficiently create new drug-like molecules for many tasks with one unified system.

In depth

The paper introduces Molexar, a unified multimodal molecular foundation model for drug design. It leverages Fragment-SELFIES, a robust, fragment-aware molecular language, and a novel value-token embedding replacement mechanism. This allows a single autoregressive decoder to handle diverse conditions—from scalar properties to protein pockets—without altering the core architecture, leading to efficient and high-quality molecule generation.

Key Takeaways

1
The authors introduce Fragment-SELFIES, a novel molecular language that explicitly encodes BRICS fragment trees, ensuring 100% validity and supporting fragment-constrained generation.
2
Molexar unifies diverse conditioning modalities (e.g., molecular properties, protein sequences, binding pockets) into a single autoregressive decoder via value-token embedding replacement, eliminating the need for task-specific architectures.
3
The model demonstrates high efficiency and performance, achieving superior validity, uniqueness, diversity, and quality in unconditional and conditional generation tasks, outperforming larger models with a smaller parameter count.

Conceptual Flow

HIGH LEVEL

Methodology: Unified Multimodal Conditioning

The model takes different types of instructions, turns them into a special code, and inserts them into a molecule's description so it can create the right kind of new molecule.

Property Goal

Protein Info

Fragment Start

Encode & Insert

Custom Molecule Request

Results: Broad Applicability and Efficiency

This new model can do many drug design jobs better and faster than older methods, creating useful molecules for various real-world challenges.

Old Separate Tools

Molexar Unified Model

Generate Drug Candidates

More Valid Molecules

Faster Design

Many Design Tasks

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.