- Case Studies
- /
- AI Generation of drug molecules with desired properties
AI Generation of drug molecules with desired properties
Multi-stage pipeline of deep neural networks that transforms discrete molecular structures into a continuous latent space
In-silico property analysis
Chemical validity
Reduction in early-stage screening time
The system allows on-demand generation and evaluation of drug candidates — entirely in silico.
Biotech
Industry
AI Software Development, Generative deep learning, Computational drug discovery
Services
The client needed an AI solution to automatically generate chemically valid drug candidates that match specified physicochemical and biological property criteria.
See what we can do for youSolution
We developed a multi-stage pipeline of deep neural networks that transforms discrete molecular structures into a continuous latent space.
Let’s talk about what’s possible
To multi-stage pipeline of several deep neural networks, Blackthorn AI applied:
Project duration
01 Month
Evaluated scientific literature and prototyped initial model structure
02-03 Month
Built and trained multiple neural networks for molecule generation & control
04 Month
Created a human-readable reporting tool for molecular outputs
05 Month
Packaged and delivered the full pipeline for operational use
Team Size
Delivering Impact
96.4%
Chemical validityAcross generated SMILES (validated via RDKit)
>10,000
Unique drug-like moleculesGenerated per run (~60 seconds on GPU)
80%
ReductionIn early-stage screening time vs traditional structure-based generation
100%
In-silico predictionOf physicochemical and bioactivity markers (QED, TPSA, logS, etc.)
