Clari's Innovation: Redefining Organic Crystal Prediction

Organic crystal structure prediction (CSP) has long been a computational headache, typically consuming years of CPU time per molecule. Enter Clari, a new generative model slashing this burden to mere seconds. While OXtal, a previous model, made strides by eliminating explicit lattice parametrization, it still demanded minutes per molecule. Clari takes that challenge head-on with a radical speed boost of 15-30 times over OXtal's capabilities.

Clari's Edge

Clari’s key innovation lies in its design. By ditching the cumbersome triangle layers for pure pair-bias attention, it streamlines the process significantly. This approach not only accelerates computations but also maintains accuracy. It's a fresh take on the problem that’s long overdue. If the AI can hold a wallet, who writes the risk model? It's Clari, with its minimal data requirements, that broadens the horizon for CSP.

Clari's model requires only basic atom types and bonds, breaking free from the constraints of traditional inputs like RDKit-sanitizable molecules. This makes it versatile enough to handle complex chemistries, including fullerenes and metal complexes. When generating 150 crystals and selecting the top 30 by energy, Clari further refines its solve rate, showcasing its prowess.

Speed and Versatility

Why does this matter? Because Clari not only improves the solve rate but also does it with astonishing speed, making large-scale virtual screening of organic solids a feasible task. While it's easy to get excited about slapping a model on a GPU rental, Clari's advancement isn’t just about speed. It's about expanding CSP's accessibility and applicability.

Clari models explicit hydrogens, allowing for inference-time scaling through direct energy ranking. This eliminates the need for any post-processing decoration or relaxation steps, adding another layer of efficiency. Show me the inference costs. Then we'll talk.

Benchmarking the Future

Clari's contributions aren't just incremental improvements. they represent a seismic shift in how we approach CSP. The introduction of the CSD Teaching Subset as a new test split sets a higher benchmark for future models. The intersection is real. Ninety percent of the projects aren't, but Clari is here to prove it's in the ten percent that matter.

In the end, this isn't just about speed or efficiency. It's about enabling practical CSP on a scale previously thought impossible. With code available on GitHub, Clari opens the door for widespread adoption and further innovation in the field.