I develop machine learning models for drug discovery applications. I am the lead developer of our generative design software, and I also work on automated data curation and predictive model building. I manage projects with multibillion-dollar clients as well as grant-funded research projects.Software Development - Lead developer of MegaSyn (an LSTM-based neural network), our generative design software. I have developed a training algorithm which allows MegaSyn to generate realistic, synthesizable molecules which simultaneously achieve multiple property goals (solubility, target activity, etc.). - Software development for automated data curation and model building, including novel neural network architectures (e.g. graph-convolutional neural networks). - I also write software as a means to extend our expertise as a company into novel areas (e.g. polymer chemistry, mass spectrometry, ionic liquids).Project Management - Manage client projects, including projects with multibillion-dollar clients. Client projects typically involve data acquisition and/or curation, predictive model building, generative design of molecules/polymers, regular meetings for updates and feedback, and writing reports for clients. - Manage grant-funded research projects, such as an effort to discover non-hallucinogenic analogues of psychedelic compounds. - Manage projects for our interns.Collaboration - I work on a number of projects managed by others in the company. My roles in these projects often include data curation, model building, generative design of molecules, meetings and feedback, and writing reports or contributing to papers. - Many of my projects involve clients and collaborators outside the company (from industry, academia, or government). Establishing a good rapport with clients and collaborators is crucial to the success of any collaborative project.

I use and develop machine learning software for a wide variety of applications in the drug discovery field. I work on everything from data curation to building predictive models, generative design of molecules, and software development. I work on grant-funded projects as well as collaborations with multibillion-dollar clients. - Lead developer for MegaSyn, our generative design software with the capacity to generate realistic, synthesizable molecules with true multitarget optimization. - Develop custom tools for automated curation of public and private data. - Acquire data and build machine learning models to predict molecular properties (e.g. biological target activity, solubility, membrane permeability). - Develop new machine learning models (graph neural networks) and molecular representations (molecular fingerprints, graph representations). - Identify novel applications to extend our expertise beyond the drug discovery field (e.g. polymer chemistry).

I used convolutional neural networks to automate image segmentation in transmission electron micrographs. As a senior member of the group, I also took a leadership role in group meetings and I advised PhD and undergrad researchers on their research projects. One of the challenges of machine learning in scientific applications is the lack of annotated data. I approached this challenge from multiple angles. To annotate existing microscopy data, I developed a custom segmentation algorithm based on a region-growing method; and I also developed a simple physical model to connect electron micrographs to physical properties such as thickness and interfacial roughness. To generate more data, I collaborated with Berkeley Lab to create realistic micrographs from simulations, which can be used to train a neural network to interpret real microscopy data.

I used Density Functional Theory (DFT) to investigate many materials systems, such as III-nitride semiconductors, high-entropy alloys, oxides, and thermoelectric materials. My PhD thesis focused on point defects in aluminum gallium nitride (AlGaN). Beyond DFT simulations, my graduate work involved many disparate areas of study, including point defect chemistry and thermodynamics, phonon calculations in the quasi-harmonic approximation, bulk properties, stochastic models, electrostatic finite-size correction schemes, and numerical methods and modeling. - Over a dozen publications in high-profile scientific journals - Multiple cross-functional collaborations with other groups - Peer reviewed and edited research publications - Trained and mentored new researchers - Developed training tutorials, manuals, and documentation - Taught classes to undergraduate and graduate students - Developed software to automate and enhance workflows - Developed tools for plotting and analyzing data - Developed simulations of electrostatics, defect chemistry, phonons and thermodynamics - Collaborative software development via GitHub - Software validation and verification (quality assurance)