Accelerate Library Design with Lib-INVENT on Vecura
This integration allows medicinal chemists and drug discovery researchers to perform reaction-informed scaffold decoration and multi-objective molecular scoring within Vecura, eliminating the need to manage complex generative AI infrastructure.
What is Lib-INVENT?
Lib-INVENT is a reaction-based generative model designed for in silico chemical library design, specifically addressing the scaffold decoration problem. By taking a molecular scaffold with defined attachment points, it autoregressively samples chemically plausible fragments to create complete, drug-like molecules. It is a lightweight, RNN-based tool that enables medicinal chemists to enumerate diverse libraries around a fixed pharmacophore.
What can users do with Lib-INVENT on Vecura?
With Lib-INVENT on Vecura, users can:
- Generate novel, synthetically plausible molecular structures by decorating fixed scaffolds.
- Apply multi-objective scoring functions to rank and filter candidates based on QED, molecular weight, and structural alerts.
- Perform rapid, in silico lead optimization without managing complex technical dependencies.
- Evaluate candidates directly against custom-weighted physicochemical and structural-alert filters in a single, streamlined API workflow.
What the output means
The output provides a list of generated molecular structures along with their associated scaffolds, fragments, and a model-calculated negative log-likelihood (NLL) indicating the probability of the decoration. Additionally, users receive a detailed scoring breakdown for each molecule, enabling clear comparison against specific performance metrics.
This output should be used to support scientific decision making. It does not replace experimental validation.
Why this matters
In drug discovery, the ability to efficiently explore chemical space around a core pharmacophore is essential for rapid lead optimization. Lib-INVENT streamlines this process by grounding generative efforts in synthetic feasibility, ensuring that proposed molecules are not only computationally promising but also more likely to be accessible via laboratory synthesis.
By integrating this capability into a user-friendly platform, researchers can bypass the setup of complex machine learning stacks and focus directly on accelerating the discovery of high-quality drug candidates.
- Developed by: AstraZeneca's MolecularAI group
- Source: Official GitHub repository
- Reference: J. Chem. Inf. Model. 2021
Vecura で Lib-INVENT を試す。
モデルワークスペースを開き、ご自身の入力で評価を始めましょう。