MAMMAL: Multi-Modal Foundation Model for Biomedical Discovery

ArticleYoel Shoshan, Moshiko Raboh, Michal Ozery-Flato, et al.Research

Researchers from IBM and collaborating institutions introduce MAMMAL, a cross-modal foundation model tailored to drug discovery. The architecture unifies protein and antibody sequences, small molecules, and gene expression profiles into a sequence-based framework, pre-trained on two billion samples.

Key points

  1. MAMMAL (Molecular Aligned Multi-Modal Architecture and Language) represents proteins, small molecules, and transcriptomic data within a unified sequence structure.
  2. The architecture integrates encoder-only and encoder-decoder mechanisms, projecting continuous numerical values directly into the embedding space.
  3. Pre-training leveraged more than two billion samples across denoising, infilling, and interaction-prediction objectives.
  4. Across 11 drug-discovery benchmarks, MAMMAL set new state-of-the-art results on nine tasks and matched competitive baselines on two.
  5. In an antibody-antigen binding benchmark, fine-tuned MAMMAL prediction scores outperformed AlphaFold 3 confidence scores in five out of seven targets.
  6. The pre-trained weights and model code are openly hosted on Hugging Face under the name ibm/biomed.omics.bl.sm.ma-ted-458m.

Why it matters

High attrition rates in drug pipelines—where roughly 90% of candidates fail in clinical trials—make early, accurate predictive modeling essential. MAMMAL demonstrates that unifying diverse biochemical modalities into a single sequence model can improve accuracy on interaction tasks beyond specialized structural prediction models like AlphaFold 3.

Analysis

MAMMAL tackles the fragmented representation problem across structural biology and transcriptomics. By combining token-based sequence modeling with continuous scalar projections, it handles classification, regression, and generative generation in one pipeline. The open release on Hugging Face lowers barriers for academic and industrial research groups seeking to benchmark or fine-tune multi-entity biological interactions.

Mentioned links

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Link:Apodex AI

    Apodex-1.1-mini: Multimodal Agentic Model Released on Hugging Face

    Apodex AI has released Apodex-1.1-mini under the Apache-2.0 licence on Hugging Face. Based on the qwen3_5_moe architecture, the model is designed for long-horizon research and agentic workflows with native tool execution.

    AI & LLMs· Tool

  • X post:Mia

    Google Releases EmbeddingGemma 2: Unified Multimodal Vector Space

    Google has launched EmbeddingGemma 2, an open-weights model mapping text, code, images, video, and audio into a single shared vector space. The 1.49 GB model is released under the Apache 2.0 licence.

    684bookmarks37kviews

    AI & LLMs· Announcement

  • Link:ATH-MaaS

    OvisOCR2: Compact 0.8B End-to-End Model for Document Parsing

    ATH-MaaS has released OvisOCR2, a lightweight 0.8-billion parameter multimodal model designed for page-level document parsing. Built on Qwen3.5-0.8B, it converts document page images directly into Markdown containing tables, LaTeX formulas, and visual bounding boxes.

    AI & LLMs· Tool

  • Article:Nikita Pavlichenko

    JetBrains Introduces Mellum2: A 12B Mixture-of-Experts Model for Code and Text

    JetBrains has released Mellum2, a 12B-parameter Mixture-of-Experts model trained from scratch on natural language and code. Activating only 2.5B parameters per token and published under the Apache 2.0 license, it is designed for low-latency, high-throughput developer and AI workloads.

    AI & LLMs· Announcement

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.