Boltz-1, first released on November 19, 2024, by researchers at the MIT Jameel Clinic, is a cutting-edge biomolecular AI model that precisely predicts the 3D spatial structure of general-purpose biomolecular complexes composed of proteins, RNA, DNA, and small molecule ligands. Similar to how a massive generative language model understands the overall context to complete natural sentences, Boltz-1 learns a pair representation of intermolecular spatial interactions and a multi-structure alignment algorithm to organically reconstruct precise 3D spatial coordinates at the atomic level. Overcoming the technical limitations of previous models that primarily focused on predicting simple protein monomer structures, Boltz-1 is an innovative computing solution in the field of bioinformatics that can generate the precise 3D structure of multi-molecular binding structures, which are the core driving force of life, and bio-fusion complexes in just a few seconds. AlphaFold2, a representative deep learning structure prediction tool, was excellent in analyzing protein interactions but had clear limitations in predicting heteromer complexes containing small molecule drugs or nucleic acids. AlphaFold3, the next-generation model, despite its high prediction accuracy, had significant constraints in incorporating it into in-house research pipelines due to the non-disclosure of model weights and restrictions on non-commercial use during its initial release. Boltz-1 fundamentally overcomes these limitations by fully and openly releasing its open-source code and all pre-trained weights under the MIT license, which allows for commercial use. Researchers can immediately build the model into their own research infrastructure without relying on closed external web servers or limited platforms, allowing them to freely perform large-scale dataset-based structure prediction and independent fine-tuning. Biotechnology and new drug development researchers can introduce Boltz-1 to efficiently build a high-throughput virtual screening pipeline for new drug target proteins and candidate drug molecules. For example, by inputting a YAML configuration file that describes the amino acid sequence of the target protein and the SMILES structure of the new drug candidate, it can be linked to an automated multiple sequence alignment (MSA) server to output the PDB structure file and structural confidence scores (pLDDT and pTM) of the binding complex within seconds. Researchers can link the generated 3D coordinate data to Python-based bioinformatics analysis tools to quantitatively evaluate the interatomic distances of the binding pocket, the hydrogen bond formation network, and the 3D spatial consistency, thereby rapidly screening the optimal lead compound with the best binding affinity from a library of tens of thousands of compounds. Furthermore, in the next-generation RNA therapeutics and immunotherapy research fields, it can be usefully introduced in experiments to precisely estimate the RNA-protein complex binding interface by simultaneously inputting the RNA base sequence and the regulatory protein chain. Optimized for GPU-accelerated performance, it completes the inference of complex multi-chain complexes in near real-time, dramatically reducing the time and significant costs associated with traditional physical laboratory-only structure analysis (X-ray crystallography, Cryo-EM). Furthermore, integrated with a binding site exploration algorithm, it is becoming a core precision analysis platform in various advanced bio-research fields, including enzyme inhibitor design, validation of the structural consistency of immune antibodies and antigens, and structural perturbation analysis of variant proteins.

Boltz-1, first released on November 19, 2024, by researchers at the MIT Jameel Clinic, is a cutting-edge biomolecular AI model that precisely predicts the three-dimensional structure of general-purpose biomolecular complexes composed of proteins, RNA, DNA, and small molecule ligands. Similar to how a massive generative language model understands the overall context to complete natural sentences, Boltz-1 learns a map of spatial interactions between molecules (pair representation) and a multi-structure alignment algorithm to organically reconstruct atomic-level precise three-dimensional spatial coordinates.

Small Molecule Binding Structure Screening for Novel Drug Candidates: Define the target protein sequence and drug small molecule SMILES code in a YAML file, execute the command boltz predict input.yaml --usemsaserver, and select the optimal ligand based on the generated 3D complex PDB file and the pLDDT binding reliability index.