Generative Compound Design with Sammon Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The traditional drug discovery pipeline is lengthy and resource-intensive, often taking 10-20 years and costing billions of dollars, due to reliance on assays with unknown clinical relevance and lack of stringent critical assessment of information, leading to high failure rates in preclinical and clinical stages.
Innovation Solution
An advanced computer-implemented method using a generative tensorial reinforcement learning (GENTRL) model to generate novel compounds active against a biological target, involving input of the target, training with reference compounds, structure generation, prioritization, Sammon mapping, and experimental validation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional drug discovery pipeline is used, then comprehensive experimental validation is achieved, but time consumption and cost increase significantly
Solution Approach 1:
The patent applies preliminary action by performing in silico filtering, generative model screening, and computational validation of compounds before physical synthesis and biological testing. This pre-screening approach identifies promising candidates in advance, reducing the number of compounds requiring time-consuming wet lab experiments while maintaining validation quality.
Solution Approach 2:
The patent segments the traditional sequential drug discovery process into parallel computational and experimental streams. Multiple generative models operate simultaneously to generate diverse compound libraries, which are then filtered through multiple in silico assays (ADMET prediction, molecular docking, QSAR) before synthesis. This segmentation enables concurrent processing that accelerates discovery without sacrificing validation thoroughness.
2Reliability
If traditional drug discovery pipeline is used, then thorough compound testing is achieved, but resource consumption increases
Solution Approach 1:
The patent applies partial action by testing only a subset of computationally-selected compounds through expensive wet lab experiments, rather than testing all generated compounds. The in silico filters and prioritization algorithms identify the most promising candidates, allowing resources to be focused on partial testing of high-probability hits while maintaining reliable validation for those selected compounds.
Solution Approach 2:
The patent replaces mechanical/physical testing systems with computational models for initial compound evaluation. In silico ADMET prediction, molecular docking simulations, and QSAR analyses substitute for early-stage physical assays, reducing resource consumption while maintaining testing rigor. Only compounds passing computational scrutiny proceed to physical synthesis and biological testing.
3Productivity
If AI generative models are used to accelerate discovery, then productivity increases, but reliability of generated compounds decreases
Solution Approach 1:
The patent applies feedback by implementing multiple in silico validation steps that evaluate generated compounds against known pharmacological and toxicological criteria. ADMET prediction models, molecular docking scores, and QSAR analyses provide feedback on compound quality, allowing the system to identify and filter out poor candidates before synthesis. This feedback loop maintains reliability while preserving the speed benefits of AI generation.
Solution Approach 2:
The patent applies preliminary action by performing computational validation of generated compounds before physical synthesis. In silico ADMET prediction, molecular docking, and QSAR filtering are conducted in advance to assess compound quality and prioritize promising candidates. This preliminary quality assessment ensures that only validated compounds proceed to expensive wet lab experiments, maintaining reliability while enabling rapid AI-driven generation.
4Adaptability or versatility
If multiple generative models are used to increase diversity, then compound variety improves, but system complexity increases
Solution Approach 1:
The patent applies universality by using a common computational infrastructure and standardized in silico validation pipeline that serves all multiple generative models. The same ADMET prediction tools, molecular docking protocols, and QSAR filters evaluate compounds from different generative models, providing a unified assessment framework. This universal validation system manages complexity while preserving the diversity benefits of multiple models.
Data Source
AI summary
A computer-implemented method can include: receiving input of a biological target; receiving a generative model (e.g., tensorial reinforcement learning (GENTRL) model or other model) trained with reference compounds, wherein the reference compounds include: general compounds, compounds that modulate the biological target, and compounds that modulate biomolecules other than the biological target; generating structures of generated compounds with the generative model; prioritizing structures of generated compounds based on at least one criteria; processing prioritized chemical structures of the generated compounds through a Sammon mapping protocol to obtain hit structures; and providing chemical structures of the hit structures. One or more non-transitory computer readable media are provided that store instructions that in response to being executed by one or more processors, cause a computer system to perform operations, the operations comprising performing the computer methods described herein for providing chemical structure of hit structures generated by the generative model.