SBIR-STTR Award

This project bridges the understanding between physical and chemical principles and genomic/proteomic response by integrating three independent parallel toxicity prediction tools. Each uses computational neural networks (CNNs) and wavelets to rapidly and accurately make pharmaceutical/chemical toxicity predictions. A CNN-based Quantitative Structure-Activity Relationship (QSAR) module makes toxicological predictions based only on structure-activity analyses; a second CNN/wavelet module makes independent toxicogenomic predictions using microarray data; and a third CNN/wavelet module makes toxicogenomic predictions using Massively Parallel Signature Sequencing (MPSS) data. This multi-intelligent, three-module approach provides crosschecks to reduce false positives and false negatives while substantially increasing confidence in predictions relative to current computer-based toxicity prediction techniques. The resulting product could potentially become a primary tool used by (a) human health researchers, b) pharmaceutical companies for screening drugs early during development, c) companies designing/developing new chemicals and chemically treated materials, and (d) government organizations (e.g., military) for mission-related chemical deployments. Public benefits include reduced health and environmental risks (e.g., 4 out of 5 chemicals in use today have inadequate testing); reduced reliance on animal testing; and reduced time and cost required to bring new pharmaceuticals and chemicals into beneficial medical and commercial use.

Thesaurus Terms:
evaluation /testing, method development, toxicology computational neuroscience, computer data analysis, toxicant screening microarray technology, polymerase chain reaction

The high cost ($0.8 - $1.7 billion) and long time frames (-13 years) required to introduce new drugs to the market contributes substantially to spiraling health care costs and diseases persisting without effective cures. A major factor is the high attrition rate of new compounds failing due to toxicity identified years into clinical trials. This particular circumstance cost the pharmaceutical industry approximately $8 billion in 2003. In silico tools generally offer the promise of identifying toxicity issues much more rapidly than clinical methods, however, they are not sufficiently accurate for pharmaceutical companies to confidently make definitive early screening and related investment decisions. LiverTox is a highly advanced, self-learning liver toxicity prediction tool that represents a quantum leap over current in silico methods. It offers a highly innovative use of multiple analytical approaches to accurately predict the toxicity of candidate Pharmaceuticals in the liver. A differentiating capability is its self-learning computational neural networks (CNNs) and wavelets. They rapidly assimilate massive volumes of information from LiverTox's extensive, dynamic, and thoroughly reviewed databases. Initially, LiverTox will generate predictions derived from five independent CNN-based submodules; one trained in advanced computational chemistry methods to make quantitative structure activity relationship (QSAR) analyses; a second trained with microarray data; a third trained with Massively Parallel Signature Sequencing and Gene Expression (MPSS/GE) data; and fourth and fifth submodules trained with proteomics and metabolomics/metabonomics data, respectively. Challenging LiverTox with new chemical formulations triggers the five independent submodules to each make toxicity endpoint predictions drawing upon its knowledge base and its similarity analysis/fuzzy logic/statistical training. This tool's flexible, highly advanced system architecture and advanced learning capabilities using data obtained from diverse techniques enable it to rapidly digest new data, build upon new data acquisition techniques, and use prior lessons learned to achieve overall toxicity predictions with greater than 95% accuracy. LiverTox's ability to rapidly and accurately predict the toxicity of drug candidates will allow pharmaceutical companies to move from discovery to curing disease faster, at greatly reduced cost, and with less reliance on animal-based tests.

Thesaurus Terms:
artificial intelligence, chemical structure function, computer program /software, computer system design /evaluation, drug discovery /isolation, drug screening /evaluation, functional /structural genomics, hepatotoxin computer data analysis, toxicant screening microarray technology