MICHAEL W. DEEM
John W. Cox Professor
B.S., California Institute of Technology (1991)
Ph.D., University of California at Berkeley (1994)
Postdoctoral Fellow, Harvard University (1995-1996)
Awards, Honors, and Positions
Fannie and John Hertz Fellow (1991-1994); Senior Research Scientist, CuraGen Corporation (1994-1995); NSF Postdoctoral Fellowship in Chemistry (1995-1996); Assistant and tenured Associate Professor, UCLA (1996-2002); NSF CAREER Award (1997-2001); Northrop Grumman Outstanding Junior Faculty Research Award (1997); Visiting Professor, University of Amsterdam (1999); A Top 100 Young Innovator, MIT's Technology Review (November 1999) (Profile); Alfred P. Sloan Research Fellow (2000); Camille Dreyfus Teacher-Scholar Award (2002); John W. Cox Professor, Rice University (2002-); Allan P. Colburn Award (2004); Editorial Board Member, Protein Engineering, Design and Selection (2005-present); Fellow, American Institute for Medical and Biological Engineering (2005); Member, Board of Directors, Biomedical Engineering Society (2005-2008); Fellow, American Physical Society (2006); Member, Rice University Faculty Senate (2006-2009); Vaughan Lectureship, California Institute of Technology (2007); Member, Nominating Committee, Division of Biological Physics, American Physical Society (2007); Member, Board of Governors, Institute for Complex Adaptive Matter (2007-present); Fellow, Biomedical Engineering Society (2009); BMES Representative on the FASEB Publications & Communications Committee (2009-2012); Professional Progress Award (2010); Fellow, American Association for the Advancement of Science (2010); External Scientific Advisor, Princeton Physical Sciences-Oncology Center (2010-present); Associate Editor, Physical Biology (2011-present); Edith and Peter O'Donnell Award, The Academy of Medicine, Engineering & Science of Texas (2012); Founding Director, Systems, Synthetic, and Physical Biology (2012-2014, raised $0.5M seed funding); Phi Beta Kappa Visiting Scholar (2012-2013); Chair, Department of Bioengineering (2014-, raised $12M in external startup funding for new faculty); Editorial Advisory Board, Bioengineering and Translational Medicine, 2016-present.
Recent Invited Talks
Theoretical methods of statistical mechanics are developed and applied at Rice to study the collective properties of biological systems. Natural systems from our world and engineered systems from biotechnology offer a wide variety of phenomena for study. The group has developed methods to quantify vaccine effectiveness and antigenic distance for influenza, methods to sculpt the immune system to mitigate immunodominance in dengue fever, a physical theory of the competition that allows HIV to escape from the immune system, the first exact solution of a mathematical model of evolution that accounts for cross-species genetic exchange, a hierarchical approach to protein molecular evolution, a `thermodynamic' formulation of evolution, and a theory for how biological modularity spontaneously arises in an evolving system. The adaptive immune response to viruses and vaccines is studied with a variety of random energy models. Field theories are used to analyze physical theories of evolution. In the materials field, the group has developed a number of widely-used Monte Carlo methods in structure, nucleation, and function of zeolites and remains interested in these areas.
Newton's laws of biology
For the last few years, we participated in a project to find fundamental mathematical laws of biology, FunBio. We contributed the idea that changes in environmental pressure stimulate the spontaneous formation of modular structure, with a proportionality constant that depends on the ruggedness of the fitness landscape. Examples of modular partitioning of the geometry of biological space can be found in protein structure, genetics, and biological networks. One way to view our research is that we seek to explain how biology nucleated from among the many possibilities in chemistry. We have described the emergence of modular organization of biological structure as a symmetry-breaking phase transition, with modularity as the order parameter. Experimental support for this description is found in pathogen structure, metabolic networks, gene networks, and protein-protein interaction networks. Additional examples include ecological food networks, developmental pathways, physiology, social networks, and economic systems.
Theory of personalized critical care
Biological health is not a single, stable, fixed point. Rather, health reflects a rich interplay of complex dynamics. Recent observations suggest that erosion of the mechanisms which underlie natural physiological complexity may be one of the most significant damaging effects of trauma or illness. For example, loss of heart rate variability is associated with deterioration of health, and loss of natural correlations in gait is associated with postural imbalance. To preempt that erosion, there is a critical need for predictive physiology. Currently, clinical predictions are based on pattern classification. By echoing predictive meteorology-that is, the use of dense data, high-speed computing, and repeated application of simple physical laws-we will make an important step in developing a framework for predictive physiology.
Thus, we seek a novel marriage between the mathematics of fluctuations and clinical medicine. Clinically, this project's importance lies in two realms. First, the clinical value of predicting what will happen "five minutes from right now" is the key to safely managing patients in acute care settings such as the emergency department, the operating room, and the ICU. Second, there is a critical need to predict when patients might veer away from their predicted physiology, that is, become "off trajectory," so that their medical team can provide heightened surveillance and early intervention. This mathematical framework, when fully formed, may serve as part of the basis for personalized critical care.
Physical theories of pathogen evolution
Following an exact solution of a simplified Eigen model by Luca Peliti, we have mapped the classical Crow-Kimura and Eigen physical theories of evolution onto quantum field theories. In this setting, we are able to exactly solve a wide class of evolution models. These exact solutions give us a 'thermodynamic' formulation of evolution that makes precise the analogies between energy and replication rate, mutation rate and temperature, and population size and temperature. These theories have been developed for fitness landscapes with multiple peaks and extended to include recombination and horizontal gene transfer. We have shown how alphabet size affects the error threshold phase transition and extended the clasical theories to finite populations.
At the applied level, we have focused on influenza A evolution. We have developed a theory of the immune pressure on the virus, both at the sequence level and at the molecular level. We developed a method based upon sequence clustering that is able to identify an emerging influenza virus before it becomes dominant. This strain detection tool would appear to be useful for annual influenza vaccine selection.
We have recently become interested in the bacterial immune system, CRISPR. We have explained, for example, the increased diversity of the leader-proximal spacer sequences in CRISPR as resulting from antigen selection pressure.
Vaccine design: Immune response to variable or multi-strain viruses and vaccines
The Deem group has developed the generalized NK (GNK) model of protein evolution. Phil Anderson suggested in the 1980s that the fitness landscape of life may be described by a spin glass model. Kauffman and Levin introduced the NK model as an instance of the spin glass model. We pointed out that biology is modular, and so structure must be present in the theoretical description. The GNK model accounts for the formation of and interaction between secondary structures of a protein and for the presence of an active or binding site in the protein.
The GNK model was used to validate the performance of a new, hierarchical protocol for protein molecular evolution. The GNK model was used to predict the usefulness of reduced alphabets in protein evolution experiments. The GNK model was used to compute the expected efficacy of B cell vaccines, as a function of the difference between the vaccine and infecting virus. The metric for antigenic distance, p_epitope, correlates to a higher degree with H3N2 vaccine efficacy in humans than do ferret animal model studies. We explained why original antigenic sin could occur, in the first theory to explain both positive and negative potential effects of a vaccine. The GNK model was generalized to T cell immunity and parametrized on altered peptide ligand experiments. Averages and correlations in the immune response were predicted. The GNK model was used to look at autoimmune disease, describing how the immune system benefits from seemingly puzzling glassy dynamics-through alleviation of autoimmune disease. The GNK model was applied to T cell immunodominance data; the predictions were within the error bars of human clinical trial data. The GNK model was used to predict influenza H3N2 vaccine efficacy; the predictions were within 10% of human efficacy data. The GNK model was used to predict immunity in an agent-based influenza epidemiology model, predicting epidemic progression in accordance with WHO data. The theory was used to predict the impact of vaccination strategy, time of vaccination, and extent of vaccination on epidemic progression.
The GNK model was used to examine the T cell response to cancer. A mechanism was put forward to explain immunodominance in cancer vaccine studies as a result of competition for access to antigen in the lymph nodes, and a quantitative comparison of GNK model predictions to specific lysis data was made. This theory suggested the multi-site vaccination strategy for viruses, which has since been validated by new experiments. The GNK model was further used to predict tumor escape and immune elimination probabilities as a function of expressed epitopes and vaccination strategy. It was shown that allele loss is more significant than point mutation for tumor escape.
Structure, nucleation, and function of zeolites
Our group developed the DIFFAX and ZEFSA methods that are widely used in the materials area, especially to solve zeolite crystal structures from powder diffraction data. We provided the first atomistic simulations, as a step for fundamental understanding, of silica nucleation under zeolite synthesis conditions. More recently we have developed a database of predicted zeolite-like frameworks that contains greater than 4 million structures. The database has been offered as a tool for the zeolite material community, and is part of the predicted crystallographically open database (PCOD). Researchers are currently exploring the possibilities that this database provides. Structures with promising properties in carbon capture and adsorption are being identified for further modeling and targeted synthesis. We remain interested in this area of materials chemistry.