Computational toxicology aims to use rules, models and algorithms based on prior data for specific endpoints, to enable the prediction of whether a new molecule will possess similar liabilities or not. In some cases, the computational models are derived from discrete molecular endpoints while in others they are quite broad in scope. Considerable progress has been made in computational toxicology in a decade both in model development and availability such that the next generation of larger scale models will promote it to the next level and further focus in vitro and in vivo testing on verification of select predictions. Pharmaceutical, consumer products and other chemistry focused companies possess structure-activity data generated over many decades of screening that is not in the public domain, and this data is primarily only accessible to the cheminformatics experts in each company. Outside of these companies small biotechs and academics must rely on data from public databases, commercial databases and their own data. Integrating such data and processing it to build algorithms that can help with predictive models is a vast undertaking. Our recent efforts have used sources like PubChem and ChEMBL to build predictive models for different toxicity related and drug discovery endpoints. Drug companies tend to focus on target related information whereas there might be an opportunity to understand toxicity also. Our work with a consumer product company indicated the need for accessing machine learning models based on the growing public datasets could be a commercially viable product. Without massive legacy internal data many of small drug discovery and consumer product companies will have to make do with data accessible in PubChem, ChEMBL or other public databases (ToxCast, Tox21 etc) only. In this proposal we will provide toxicity machine learning models developed with different algorithms (Bayesian, Support vector machines, random forest and Deep Neural Networks as just some examples) for 40-50 in vitro and in vivo datasets. We are not aware of any other company pursuing such an approach to create as wide an array of toxicity models. We are also not aware of other software companies in the toxicity model space generating their own experimental data to test the models. Such a technology could have very broad utility for other pharmaceutical companies, biotechs, consumer product companies, regulatory groups and academic research groups. As we have been working for several years to build up technologies and experience in cheminformatics and software development it would be relatively straightforward for us to build the core foundations of MegaTox and deliver them to potential customers. This technology will also be used in our consulting projects with pharmaceutical and consumer product companies.
The objective of ?MegaTox? is to develop a new computational system and tools for integrating toxicity data, and machine learning models. This will enable the prediction of various toxicity endpoints relevant to pharmaceutical and consumer product companies. We will use the latest machine learning algorithms, access the latest curated public datasets and develop a user intuitive interface and visualization system. We will use MegaTox internally for our own drug discovery projects as well as leverage it for consulting for clients so we can combine public and private datasets and increase the coverage of chemical property space. We will perform generate toxicity data in house and with collaborators to test our models.
