This Small Business Innovation Research (SBIR) Phase I project proposes to develop a web-based learning software platform for collaborative drug discovery and optimization. Current drug discovery software tools do not support the sophisticated data management that is needed for iterative and adaptive drug discovery, learning, and collaboration. This project aims to change this by building a web-based drug discovery platform with three main objectives: (1) provide novel insightful graphical feedback on binding properties, (2) provide simple virtual screening over millions of vendor compounds, and (3) keep track of past assay results in order to iteratively refine prediction models and to continuously improve accuracy. The challenges in this project include (1) scalable data management to support hundreds of users and large numbers of compounds and models, (2) data encryption for the protection of proprietary information, and (3) a simple pharmacologist-oriented user interface. The outcome of the project will be a web-based tool that can be used across pharma/biotech research groups and across labs for effective collaborative drug discovery and optimization.
The broader impact/commercial potential of this project, if successful, will be the potential for novel visualizations that will significantly improve the understanding of molecular binding properties. It will result in novel data mining and analysis algorithms that can operate directly on encrypted data, with far-reaching impact on other areas that require data safe-keeping (e.g., electronic health record management). In addition, the project will offer insights into interface design for pharmacologists, biologists, and chemists and how to better foster collaboration among them. On a larger scale, the project will aid in the discovery of novel leads for pressing healthcare needs. Especially for difficult drug targets without much structural information, such as Alzheimer's disease targets or cancer targets, the proposed platform can identify novel drug-leads with fewer side effects and higher efficacy. This ultimately may result in a much faster time-to-market for such drugs. For pharma/biotech companies using the platform developed in this project, this could mean a significant commercial advantage.
This Small Business Innovation Research Phase I project investigated how to make collaborative drug discovery available in the web. More specifically, we designed and developed a prototype drug discovery software-as-a-service (SaaS) platform termed "LEADS" that is tailored to a user's drug discovery project and that improves over time with additional assay data becoming available. Activity prediction is of immense importance to the pharmaceutical industry as it allows evaluating biological properties early on in the drug development process. Current approaches analyze compounds on a target-by-target basis. LEADS' technology can explore multiple targets at once by operating in a joint pharmacophore space of chemical compounds, targets, and physiochemical/biological properties. At a high level, the proposed LEADS platform operates as follows. A user can upload measurement results from assays (both target-specific or biological). These results are securely stored and can be shared with collaborators. Users can then build models for the observed assay activities. These models are based on Acelot's JPS (Joint Pharmacophore Space) technology that analyzes relevant points (the "pharmacophore") of all assayed molecules to identify significant pharmacophore constellations. Once the models are built, they can be shared with colleagues and the discovered pharmacophore hypotheses can be visually inspected. The models can then be used to screen a large database of molecules to find other novel candidates with the same properties. When more assay data becomes available, it can be added to the JPS model to incrementally improve the predictions and tailor them more to the drug discovery project. Since LEADS is fully web-based, users do not need to install or maintain any software or compute clusters. In this project, we pursued three objectives: firstly, we tuned the underlying machine learning algorithms and added confidence intervals to the predictions. This provides more meaningful outputs to the chemist or pharmacologist using LEADS. In addition, a web-based 3D pharmacophore visualization was added that allows users to visually inspect the discovered pharmacophore constellations. Secondly, a large database of drug-like compounds for virtual screening was created from many compound vendors. This database currently contains 20 million compounds and can be searched by LEADS to find molecules with desired properties, based on the built JPS models. Thirdly, all components of LEADS were integrated in an SaaS platform with a web-based user interface (see LEADS screen shot). This objective also included the development of easy-to-use components for uploading assay results, for JPS model building, for starting virtual screening runs on Acelot's compute server, and for delivering screening results to the user. We believe that a shared, web-based platform like LEADS can drastically change the way drug discovery occurs with more insightful analyses, better data provenance tracking, and less software maintenance issues. The incremental model building and model/assay data sharing will lead to an increased rate of high quality lead candidates discovered. LEADS can also further our understanding of basic compound properties, such as blood-brain barrier permeability and biological pathway interactions, provide insights into the interaction of ligands and proteins, and will increase collaboration between researchers. This will aid in reducing the cost for drug development (and thereby for the final drugs themselves), lead to shorter drug development cycles and time-to-market, and lead to novel discoveries and treatment options.
