SYNCOGEN: A New Machine Learning Framework for Generating 3D Molecules That Are Both Structurally Accurate and Synthesizable

Scale AI, a leading data-labeling startup, has confirmed a major investment from Meta, boosting the company’s valuation to $29 billion. As part of the deal, Scale’s co-founder and CEO Alexandr Wang will step down and join Meta to support its efforts in building superintelligent AI systems. The investment, reported to be around $14.3 billion for a 49% stake, underscores Meta’s strategic push to strengthen its AI capabilities amid competition from rivals like OpenAI and Google. Wang’s departure follows a trend of top AI talent leaving startups for larger firms, as Meta seeks to accelerate its generative AI development. Jason Droege, Scale’s current chief strategy officer, will serve as interim CEO, with Wang remaining on the board. The funding will be used to return capital to shareholders and drive growth, though Scale emphasized its independence. The move highlights the growing importance of high-quality training data in the AI race, as companies like Meta invest heavily in infrastructure to stay competitive. SYNCOGEN, a machine learning framework developed by researchers from the University of Toronto, University of Cambridge, and McGill University, addresses a critical gap in drug discovery by generating 3D molecular structures that are both chemically realistic and synthetically feasible. Traditional AI models often produce molecules with desirable properties but lack practical synthetic routes, limiting their use in pharmaceutical research. SYNCOGEN overcomes this by jointly modeling reaction pathways and atomic coordinates, ensuring generated molecules can be constructed from existing building blocks using known chemical reactions. This dual approach bridges the divide between computational design and laboratory implementation, a major hurdle in AI-driven drug development. The framework leverages the SYNSPACE dataset, which contains over 600,000 synthesizable molecules derived from 93 commercial building blocks and 19 reaction templates. Each molecule is paired with multiple energy-minimized 3D conformations, totaling 3.3 million structures, to provide a robust training resource. SYNCOGEN’s architecture builds on SEMLAFLOW, an SE(3)-equivariant neural network optimized for 3D molecular generation. It combines masked graph diffusion for reaction pathways with flow matching for atomic coordinates, incorporating constraints like edge count limits and compatibility masking to enforce chemical validity. During training, the model uses graph cross-entropy, coordinate mean squared error, and pairwise distance penalties to balance structural accuracy and synthetic tractability. Benchmarking shows SYNCOGEN outperforms existing all-atom and graph-based generative models in 3D molecule design, achieving state-of-the-art results. It excels in molecular inpainting for fragment linking, a key task in drug design, generating analogs of complex molecules with favorable docking scores and retrosynthetic pathways. This capability is a significant advancement over conventional models, which often prioritize 3D geometry without considering synthesis feasibility. The framework’s potential extends to property-conditioned generation, protein binding pocket targeting, and integration with lab robotics for automated synthesis. Industry experts highlight SYNCOGEN’s role in transforming computational chemistry by aligning AI-generated molecules with real-world lab constraints. Its dataset and architecture could become foundational tools for pharmaceutical research, reducing the time and cost of drug development. The collaboration between academic institutions and industry leaders like Meta reflects a growing trend of cross-sector innovation in AI-driven science. Researchers note that SYNCOGEN’s focus on synthetic accessibility addresses a long-standing challenge, making it a critical step toward practical AI applications in chemistry. The project’s open-source nature and detailed documentation position it as a valuable resource for the scientific community, with implications for materials science and beyond.