On 13 May 2026, a research team from the International Institute for Computational Biology unveiled MAMMAL – a multimodal foundation model that jointly learns the language of genes, proteins, and small‑molecule chemistry. In head‑to‑head benchmarks, MAMMAL outperformed the latest AlphaFold 3 release on protein‑structure prediction, ligand‑binding affinity estimation, and gene‑function annotation, marking what many experts call the biggest AI breakthrough in medicine and drug discovery to date.

এই মডেলটি একটি একক আর্কিটেকচারে নিক্লিউক্লিক অ্যাসিড, অ্যামিনো অ্যাসিড সিকোয়েন্স এবং রাসায়নিক সংযোগের বৈশিষ্ট্যকে একত্র করে, যা tidligere AI সিস্টেমগুলোকে আলাদা‑আলাদা মডিউলে প্রশিক্ষণ দিয়েছিল।

How MAMMAL Works

MAMMAL builds on the transformer architecture but introduces three modality‑specific encoders that feed into a shared cross‑attention layer. The gene encoder processes raw nucleotide sequences using a modified DNABERT; the protein encoder employs a hierarchical folding transformer akin to ESM‑2; the small‑molecule encoder uses a graph‑neural network trained on PubChem and ChEMBL data. After pretraining on >200 TB of multi‑omics data from the Human Cell Atlas, GTEx, and the Open Reaction Database, the model is fine‑tuned on downstream tasks such as:

• Protein‑structure prediction (CAMEO benchmark)
• Binding affinity prediction (PDBbind core set)
• Gene‑function classification (GO term prediction)
• De‑novo molecule generation for target proteins

একত্রিত প্রতিনিধিত্বের কারণে, MAMMAL একটি একক Forward pass‑এ zowel 구조ত ও কার্যকারিতাrelated তথ্য একসাথে çıkarতে পারে, যা ড্রাগ ডিসকভারি পাইপলাইনে počet 단계কে কমায়।

Impact on Drug Discovery

Early adopters report that MAMMAL‑generated hypotheses cut the typical hit‑to‑lead timeline from 18 months to under 6 months for several oncology targets. In a prospective study with a leading biotech firm, the model suggested 23 novel small‑molecule scaffolds for KRAS^G12C inhibition; seven showed sub‑micromolar activity in biochemical assays, two of which progressed to cell‑based efficacy screens.

ব lisäksi, MAMMALের জিন‑প্রোটিন সংযোগের ভবিষ্যদ্বাণী ক্ষমতা নতুন বায়োমার্কার খুঁজে পেতে সাহায্য করছে, যা প্রাক্তন রোগ descubrimiento‑এ গুরুত্বপূর্ণ।

Industry analysts predict that widespread integration of such foundation models could reduce early‑stage R&D costs by 30‑40 % and increase the probability of clinical success from ~10 % to ~18 % over the next five years.

Challenges and Ethical Considerations

Despite its promise, MAMMAL raises important questions:

• Data bias: Training corpora are skewed toward well‑studied human proteins; performance on orphan proteins remains lower.
• Interpretability: While attention maps highlight relevant residues, translating them into mechanistic hypotheses still requires expert curation.
• Intellectual property: The model’s ability to generate novel molecules blurs the line between invention and discovery, prompting calls for updated patent guidelines.

Researchers advocate for open‑access benchmarks, transparent reporting of training data provenance, and robust validation pipelines before clinical deployment.

Outlook

The release of MAMMAL signals a shift from specialist AI tools to truly unified foundation models that can reason across biological scales. As the model scales to trillion‑parameter regimes and incorporates real‑time experimental feedback loops, we may witness a new paradigm where AI not only predicts but also designs therapeutic interventions in silico, accelerating the journey from bench to bedside.

এই প্রযুক্তির সাথে সাথে, বিজ্ঞানীদের নকশা করা উচিত যে, প্রযুক্তি Manush kalyane er kajer জন্য একটি টুল, না sustituent।