Deep Learning for Biological Data Integration
Most machine and deep learning methods for biological data process data elements in isolation, rather than in-context, limiting their performance and applicability for understand complex ecological systems. Data such as high-throughput sequencing, gas/liquid chromatography mass spectrometry, and transcriptomics are also commonly collected, but are typically analyzed independently and connections between data domains are often performed post hoc (or not at all). We address these concerns by developing/training/testing deep learning architectures capable of reading complex biological data and are not limited to single sequence or single domain analyses. We take hints from recent developments if AI where human communication data (text, images, audio) is used to produce foundation models using multimodal pretraining and fine-tuning methods in order to provide multimodal pretraining and fine-tuning methods for foundation models for biological data. Relevant publications: [Lugwig et al., 2025] [Alexander et al., 2025] [ Romer et al., 2024]
Accelerated Molecular Dynamics Simulations
Molecular dynamics simulation remains the gold standard simulation paradigm for biomolecular modeling due to the careful balance between physical accuracy and computational efficiency for atomic-scale phenomena. However, short simulation timescales still prohibit its use for many biophysical phenomena of interest. Metadynamics methods often employ explicit bias in either the potential or kinetic energy to overcome this limitation, but the impacts of such bias on the results are not easy to interpret and may potentially hinder correctness. We explored the adaptation of traditional search algorithms from AI to molecular simulations and were able to reduce the computational overhead needed to produce guided molecular dynamics simulations by several orders of magnitude witout introducing artifical bias in either the potential or kinetic energy landscape. Relevant publications: [Syzonenko and Phillips, 2020]
Environmental Effects on Protein-Protein Interactions
A complete understanding of how environmental factors influence HIV transmission is needed for the development of an effective HIV vaccine. I develop and utilize state-of-the-art modelling and simulation methods to investigate protein-protein interactions involved in viral transmission at the molecular level, and determine how relevant environmental factors (such as salt concentration and pH) affect these interactions. This project was a collaboration with researchers at Los Alamos National Laboratory. Relevant publications: [Morton and Phillips, 2021] [Morton and Phillips, 2021] [Morton, Phillips and Phillips, 2019] [Morton, Howton and Phillips, 2018] [Gottardo et al., 2013] [Stieh et al., 2013]
Mutational Analysis of Metalloproteins
Modelling and simulation provide a rational approach for predicting how mutations to metalloprotein enzymes will degrade or enhance catalysis of viable substrates. These predictions greatly aid the search for useful mutations by prioritizing the screening order of mutants in the laboratory, lowering costs and producton times. Key targets have been enzymes which degrade chemical warfare nerve agents and may be used for therapeutic or sanitation purposes. This project was a collaboration with researchers at Los Alamos National Laboratory.
Multiscale Modeling and Simulation of Multidrug Resistance
Bacteria have developed several mechanisms which contribute to multidrug resistance and hinder the development of new antibiotic treatments. If the mechanisms involved in resistance were more fully understood, novel treatments might be developed which could rescue our current stream of antibiotics. I use modelling and simulation to study the structure and function of multidrug resistance efflux pumps, one of the main contributors to antibiotic resistance. I aim to uncover a complete mechanistic understanding of pump function in order to aid efforts to effectively subvert their function. This project was a collaboration with researchers at Los Alamos National Laboratory. Relevant publications: [Phillips and Gnanakaran, 2015]
Molecular Simulation Analysis
There are many biological and chemical processes which can be studied in remarkable detail using computational modeling and simulation. The significant computational cost of these approaches suggests that rigorous analysis of the resulting data is necessary to justify the consumed computational resources, and state-of-the-art statistical and machine learning methods are poised to fill this need. However, these methods are often developed outside of the scientific domains where they are applied, and the nuances faced when working with real data make it difficult to discern when these approaches are achieving their intended purpose. I use domain knowledge to construct model-based validation frameworks which help to resolve such issues. Past domains of interest include intrinsically disordered and natively folded proteins. This project was a collaboration with researchers at the University of California, Merced. Relevant publications: [Syzonenko and Phillips, 2018] [Phillips, Colvin and Newsam, 2018] [Phillips, Colvin and Newsam, 2011] [Yamada et al., 2010] [Phillips, Colvin, Lau, and Newsam, 2008]
Robot Prefrontal Cortex (PFC) Working Memory Toolkit (WMtk)
One past project focused on the development of biologically inspired computational mechanisms for effective robot learning and control. In particular, David Noelle (Univ. of Calif., Merced), and I developed a software toolkit that allows for the easy integration of a powerful computational neuroscience model of working memory into robotic systems. Current work involves combining the toolkit with models of other brain systems and creating more efficient knowledge representation structures which are more flexible and comprehensive than those currently used by toolkit. This model of working memory has been used to train robots to perform standard laboratory tests of working memory function, such as the delayed saccade task, as well tasks in robot navigation, motor skill learning, and object manipulation. More recent work involved adapting the framework to modern deep learning frameworks and tasks. Relevant publications: [Miller, Naderi, Mullinax, and Phillips, 2022] [Ludwig, Remedios, and Phillips, 2021] [Omatu and Phillips, 2021] [Khan and Phillips, 2020] [Williams and Phillips, 2020] [Jovanovich and Phillips, 2018] [Williams and Phillips, 2018] [DuBois and Phillips, 2017] [Phillips and Noelle, 2005]