Computer-Aided Drug Design (AIDD)
Computational chemistry services that narrow the synthesis list before a single flask is used. Structure-based design, virtual screening, free-energy methods and machine-learning models run alongside our medicinal chemists, so predictions are tested in weeks rather than filed away.
AI is only useful in drug discovery when it changes what gets made. Our AIDD group sits inside the chemistry organization, not beside it — generative models, virtual screening, and ADMET prediction feed directly into the medicinal chemistry cycle, so computational output becomes synthesized compounds and real data rather than a report. AIDD can join a program at any stage, from target identification through lead optimization.
Our platform spans the full computational workflow. Target discovery works bidirectionally — elucidating mechanism of action, identifying and confirming targets, exploring repurposing opportunities, and predicting off-target and toxicity risk. Structure prediction uses AlphaFold-based deep neural network models to resolve target protein structures where experimental structures are unavailable. De novo design draws on more than ten generative models to produce complete molecules or modified scaffolds, enabling scaffold hopping around existing patent space. Virtual screening covers an in-stock library of over 12 million compounds, filtered for structural novelty against known patents and scored through combined AI and CADD approaches. And our ADMET platform predicts more than 160 drug-likeness parameters — including Caco-2 permeability, human and rat liver microsomal stability, and phototoxicity — using proprietary models, with the option to build custom models on your own data.
Beyond small molecules, we apply the same approach to antibody design (epitope analysis, CDR-focused affinity maturation, developability prediction), ADC linker design through Linker-GPT, our transformer-based generative model fine-tuned on 3,700+ known ADC linkers, payload design via pocket-based diffusion models with multi-objective optimization across potency, toxicity, and polarity, and degron discovery for molecular glue programs.
What It Changes
In a recent de novo design program, we started from 574 patent compounds and generated 78,011 novel structures. Successive rounds of docking, drug-likeness and ADMET filtering, and FEP with expert medicinal chemistry validation narrowed that to five compounds for synthesis. Three of the five reached potency close to the reference compound — with a clean IP position.
The comparison that matters is against how the same work is usually done. A traditional lead optimization campaign of this type takes three to five medicinal chemists, 1.5 to 2 years, and roughly $1.0–1.5M. Ours took one to three chemists, eight months, and roughly $0.5–1M — better than a 50% reduction in both time and cost. Across our project history, AIDD has supported programs in CNS, oncology, metabolic disease, and cardiovascular indications for pharma, biotech, and academic partners, spanning target identification through IND.
