The Electrical Engineering and Computer Science Department at MIT offers its Ph.D. students the option to specialize in research at the interface between Computer Science and the Life Sciences. Our long heritage of research in this area includes, among other areas, work in computational biology, synthetic biology, medical image analysis, bioinformatics, and computer assisted surgery. Students interested in research in these areas are encouraged to review the brief descriptions of some of the current activities of faculty and staff in Computer Science that relate to the life sciences, listed below by faculty member. In addition, the Electrical Engineering and Computer Science Department has an NIH Training Grant to facilitate support for graduate students who are doing innovative work in the Life Sciences, broadly construed. Many life sciences faculty members are involved in this training program that provides an interdisciplinary framework for Ph.D. research. Additionally, many life sciences faculty members are engaged with other collaborative projects at MIT, links to which are provided below.

Research in bioengineering at the interface with electrical engineering is also active within the department. Further information can be found at (Bioelectrical Engineering, Area VII).

Faculty and Staff

· Robert Berwick

· Luca Daniel

· Eric Demaine

· David Gifford

· Polina Golland

· Eric Grimson

· John Guttag

· Tommi Jaakkola

· Manolis Kellis

· Tom Knight

· Tomas Lozano-Perez

· Steve Massaquoi

· Collin Stultz

· Peter Szolovits

· Bruce Tidor

· Jacob White

Training Grant

The EECS Department has recently secured an NIH training grant, which provides opportunities to support students working in computational life sciences.

Other affiliated programs

In addition to individual faculty projects, many faculty members are actively engaged in collaborative projects with other life science programs affiliated with MIT. These include:

· The Eli and Edythe L. Broad Institute – genomic medicine

· The Harvard-MIT Division of Health Sciences and Technology (HST) – biomedical imaging; biomedical informatics and integrative biology, and regenerative and functional biomedical technologies

· The MIT Computational and System Biology Initiative (CSBi) – quantitative models of complex biological systems, with application to medical or pharmaceutical research

· MIT Biological Engineering Division – application of engineering methods to biological problems

· MGH/MIT/HMS Athinoula A. Martinos Center for Biomedical Imaging – biomedical imaging and image analysis

Robert Berwick

The goal of Professor Berwick’s work is to use applied mathematical and modeling principles to develop abstract models for complex biological systems. Research is proceeding in two areas: 1) understanding the relationship among genotype, environment and phenotype; and 2) developing a synthetic approach to modeling biological systems.

Relationship among genotype, environment and phenotype
Information about a simple gene sequence does not necessarily predict protein function. Work with in vitro expression and recombinant protein production systems has illustrated that post-translational modification and environmental factors can influence how a primary amino acid sequence folds in three-dimensional space to perform a particular biochemical or physiological function. These interactions cannot always be deduced from primary nucleotide sequence information by reductionist approaches. We are developing general, thorough mathematical models that unwind the complexities in these systems and that describe the interrelationships among genes, their cellular environment, and the functional proteins that they produce.

Synthetic approach to modeling biological systems
Much of the current modeling work takes a pragmatic reductionist approach to complex biological systems, paring away layers of detail and focusing on single pathways that are more tractable to experimental manipulation and to simulation. Now we have sufficient data and modeling tools to put these systems back together into a single unified model that incorporates multiple pathways and their environmental interactions.
One aspect of our project compares simulations of complex systems with experimental data derived from manipulations of individual components. We have used a modified stochastic Petri net model to describe the entire apoptosis process, from the initial triggering events through DNA fragmentation and cell death. This system uses qualitative experimental information about biochemical interactions and incorporates more than 300 individual elements, dozens of pathways, several feedback loops, and intracellular organelles such as mitochondria. It allows us to track the flow of individual molecules among the nodes in a signal network, and to compare predictions about DNA fragmentation with experimental data derived from individual genes, pathways and feedback loops.
Another aspect of our modeling project focuses on how biological systems evolve over time. Decision points that are crossed during the natural evolutionary process appear to limit the choices that can be made later on. For example, all vertebrates that developed wings to fly did so by sacrificing two existing limbs, but insects were able to add a body segment and develop a supplementary set of limbs that became wings. We are intrigued by this type of evolutionary limitation and want to understand its origin. Two interesting applications of this work are to predict the phenotypic changes that we might expect from specific genetic modifications that we can make in complex living systems, and to develop computational methods for high-throughput in vitro toxicology screening.

Erik Demaine

Professor Demaine does his research in the Computer Science and Artificial Intelligence Laboratory, where he is a member of the Theory of Computation group. Professor Demaine's research centers around algorithms, data structures, and geometry.

Professor Demaine's research in computational biology focuses on mathematical and computational models of protein folding and design. These models range from the basic kinematics of fixed-angle linkages, to the hydrophobic-hydrophilic energy function.

For more information, see Professor Demaine's webpage.

David Gifford

Our laboratory develops new machine learning techniques and algorithms to model the transcriptional regulatory networks that control gene expression programs in living cells. We have a very productive interdisciplinary collaboration with leading biologists that has allowed us to tackle extraordinarily difficult and interesting problems that underlie cellular function and development. For example, we have developed probabilistic models of cellular function (PSB), built a comprehensive model of the yeast cell cycle (Science 2002), participated in the discovery of the draft transcriptional regulatory code of yeast (Nature 2004), and helped uncover how key diabetes related transcription factors regulate cellular function in the human pancreas and liver (Science 2004). Current work in our laboratory is examining how we can computationally model chromatin modifying complexes that are associated with the genome of living yeast cells. New kinds of mechanistic computational models are necessary to capture how chromatin structure encodes cellular memory, and how the state of this memory is used to control gene expression. In particular, we are investigating new modular graphical models that use mechanistic constraints to describe biological mechanism.

A new focus is an interdisciplinary project that seeks to build computational models of the transcriptional regulatory networks that control the differentiation of specific cell types. Elucidating these regulatory networks will enable us to define the regulatory processes that determine a cell's progress to its terminally differentiated state, and position us to differentiate embryonic stem (ES) cells for the treatment of debilitating human diseases. New computational techniques for elucidating transcriptional regulatory networks based on the integration of diverse high-throughput experimental data (genome sequence, chromatin structure, transcription factor-DNA binding, gene expression) provide a powerful foundation for discovering the detailed mechanisms of regulatory network control of cell differentiation during development.

Polina Golland

Our group focuses on computational methods for characterization of biological processes using images as a source of information. This research straddles the boundary between Computer Vision and Machine Learning in application to medical and biological domains.

Shape Analysis: We are developing techniques for characterizing biological shape and its variability, for detection of changes in the shape and appearance when a population of organisms is observed. Such methods are important in many applications, from modeling cortical folding in neonatal brains (using MRI for imaging) to understanding how genetic perturbations affect appearance and behavior of cells (based on high throughput microscopy images).

Function Analysis: Functional MRI (fMRI) enables non-invasive observation of the brain activity, but its relatively low time resolution and high noise make robust detection of the signal very difficult. We are working on improving the detection quality, as well as characterizing the function of the brain as measured by fMRI at the system level, building models that reflect which components act in coordination and how this coordination is affected by different diseases.

For more information, please see Golland.

Eric Grimson

Professor Grimson’s group focuses on the application of computer vision and machine learning methods to problems of medical image analysis. Current work focuses on image guided surgical interventions, and computational anatomy.

Image guided surgical interventions deals with the problems of creating patient-specific, detailed reconstructions of anatomy from multi-modal medical images, and registering those models to the patient during surgical procedures. Such models enable surgeons to plan procedures effectively, and registration of the models to the patient supports surgical guidance and navigation. Key technical problems include multi-modal registration, adaptive registration, segmentation of images into tissue classes based on intensity, anatomy and shape information, and visualization of these models. Applications of the methods include neurosurgery, prostate surgery, orthopedic surgery, ENT surgery.

Computational anatomy focuses on analysis of organ structure and substructures, across time and across populations of patients. Of particular interest is extracting representations of shapes of structures that support statistical characterization. These models can be used to develop classifiers for identifying new examples, and for isolating significant differences in shape and connectivity between populations. Of particular interest are the use of such shape models in the study of schizophrenia, Alzheimer’s disease, and other neurological disorders.

For more information, please see (Medical Vision Group). We have collaborated extensively for over a decade with the Surgical Planning Laboratory at Brigham and Women’s Hospital and Harvard Medical School.

John Guttag

Pervasive Physiological State Sensing and Decision Making
Medicine is a high tech profession, yet the rate of innovation has been slow compared to that of other high tech areas. Certainly, some amount of inertia is required to insure the necessary level of safety, but in many cases the slow rate of innovation has little to do with safety concerns. There is an abundance of physiological information that could be made available to the clinician, but most of the “sensor fusion” is done in the mind of the caregiver, whose availability, memory, and computational capacity is limited. Our group develops technologies and systems to make the practice of medicine more effective, safer, and more efficient. These technologies are designed to be effective in both clinical and non-clinical settings.

We are currently focused on developing systems to help in the monitoring and management of chronic conditions. These systems will, in the fullness of time,

• Use a variety of physiological sensors to monitor subjects on a continuous or intermittent basis,
• Use sophisticated signal processing techniques to extract relevant features from individual and fused signals,
• Store both raw signals and derived information on permanent storage,
• Incrementally build a subject-specific physiological model,
• Use the model to aid in medical decision making, and
• Initiate activities that will lead to therapeutic actions.

Tommi Jaakkola

Prof. Jaakkola's group focuses on several complementary research areas: statistical inference and estimation, machine learning, and computational biology. In the area of computational biology the motivation comes from the need to understand cellular mechanisms responsible for transcriptional control. This is a problem of enormous scientific and practical importance. Our work has focused first on model organisms such as yeast with the ultimate goal of understanding regulatory control in more complex human cells. Specific research topics include computational analysis of biological processes, large scale reconstruction of molecular interaction networks, and automated selection of new informative experiments to be carried out.

For more information, please see (Jaakkola).

Manolis Kellis

Professor Kellis’ group works in the area of Computational Biology, pursuing the understanding of biological systems using algorithms and machine learning. Specific areas include genome interpretation, comparative genomics, regulatory networks, cellular signals, developmental biology, and evolutionary theory.

(1) in the area of genome interpretation, they are developing comparative genomics methods to identify genes and regulatory elements in the human genome

(2) in the area of gene regulation, they are studying the combinatorial control of gene expression and cell fate specification

(3) in the area of evolutionary genomics, they study the emergence of new gene functions, turnover of regulatory motifs, and the coordinated evolution of functionally interconnected cellular components.

For details, please visit KELLIS.

Tom Knight

Much of the work at the intersection of biology and EECS is devoted to the tools, instrumentation, and software of use to the biology community. Our research, in contrast, focuses on the development of biology as an engineering discipline — a field we term synthetic biology. In the last century, we molded the science of physics into powerful engineering technologies using our tools of abstraction, parts, isolation, black boxing, feedback, and hierarchy. Dramatic advances in the science of biology enable us to leverage these same tools to create a powerful engineering discipline of synthetic biology, in which the design and construction of functional, artificial, living systems is the explicit goal.

Our laboratory, in close collaboration with Prof. Drew Endy in the Bioengineering Division, is working to develop a kit of standard biological components, the means of assembling those components, and the tools for characterizing, and modeling them. Circuit design of genetic systems is in its infancy, but will be an important component of our work. Similarly, the complexity of even simple biological organisms is almost overwhelming. We seek to reverse engineer, simplify, and rebuild the cells of very simple bacterial species, creating an engineered cell suitable for use as the chassis and power supply for our circuits.

For more information, please see (synthbio) or (parts).

Steve G. Massaquoi

Our group seeks to develop engineering models of human movement control systems. The purposes are to a) increase scientific understanding of normal and pathological motor function, b) provide a foundation for advanced diagnostic and therapeutic interventions in neurological disorders (e.g. neuroprosthetics), c) provide a foundation for the design of biomimetic artificial systems, e.g. humanoid robots. Emphasis is placed on accurate characterization of nervous system function at the level of functional neural assemblies, reproduction of the firing activity waveforms of these units, and engineering analysis of system performance. Theoretical formulations are tested by extensive computer simulations and human experiments, involving both healthy subjects and those with neurological disorders, using laptop-based tests and/or a robotic manipulandum. Engineering issues include the achievement of stable control in presence of neural processing delays, adaptation of low-level kinematic and dynamic control in free space and during environmental contact, hierarchical and horizontal distribution of control system design, robustness of system performance in presence of noise and neuronal loss, motor programming/behavioral learning. Areas of physiological interest include spinal cord function and motor synergies, cerebrocerebellar interaction in adaptive coordination and dynamic compensation, fronto-basal ganglionic interaction in motor learning and programming.

Peter Szolovits

The Clinical Decision Making Group conducts research on computer modeling of medical knowledge, methods of supporting clinicians and patients in making medical decisions, monitoring systems that integrate multiple types of clinical data, systems for collecting and maintaining life-long medical records that are controlled by patients, language understanding methods that help extract meaningful knowledge from often poorly-organized textual medical records, home-based health care, and improvements to the confidentiality and security of medical data. We also work on integrating the expanding amount of accessible biological data with the clinical records that are our best approximation to human phenotypes, to help bridge from the research laboratory to the bedside.

For more details, please see MEDG.

Collin Stultz

Laboratory of Computational Biophysics and Molecular Engineering: Research in the Stultz group is focused on understanding conformational changes in biomolecules that play an important role in common human diseases. Since conformational transitions in biomolecules are typically difficult to observe experimentally, we use novel methods to gain insights into the role that molecular structure plays in the progression of human disease. The lab uses an interdisciplinary approach combining computational modeling with biochemical experiments to make connections between conformational changes in macromolecules and disease progression. By employing two types of modeling, molecular dynamics simulations and probabilistic modeling, hypotheses can be developed and then tested experimentally.

For more information see http://csbi.mit.edu/faculty/Members/cmstultz

Bruce Tidor

Research in the Tidor Group is focused on the analysis of complex biological systems at the molecular and network levels. Projects at the molecular level study the structure and properties of proteins, nucleic acids, and their complexes. Investigations probe the sources of stability and specificity that drive macromolecular folding, binding, and catalysis. Studies are aimed at dissecting the interactions responsible for the specific structure of folded proteins and the binding geometry of molecular complexes. The roles played by salt bridges, hydrogen bonds, side-chain packing, rotameric states, solvation, and the hydrophobic effect in native biomolecules are being explored, and strategies for re-casting these roles through structure-based molecular design are being developed. Work at the network level involves the study of biochemical regulatory networks and signal transduction pathways in cells. The development of approaches to relate network topology to functional characteristics is fundamental to this research. Significant effort is being applied to extracting the design principles for biological networks and to understanding the control functions implemented. The insights resulting from this work will provide a strong foundation for understanding biological systems; moreover, they will be useful for the development of therapies that ameliorate disease states, as well as for the construction of new synthetic systems from biological components. The methods of theoretical and computational biophysics and approaches from computer science, artificial intelligence, applied mathematics, and chemical and electrical engineering play fundamental roles in this work.

For more information, please see TIDOR.

Jacob White and Luca Daniel

Research in the computational prototyping research group (Professors L. Daniel and J. White) uses a range of engineering design applications to drive research in numerical simulation and optimization algorithms and software, with an emphasis on developing novel computational techniques for application problems which have heretofore been considered intractable. Our current efforts are focused on several application areas including: integrated circuit and package design, micromachined device development for biological applications (BIOMEMS) (in collaboration with Professors Voldmann and Han), aircraft analysis (in collaboration with Professor Peraire), protein optimization for drug synthesis (in collaboration with Professor Tidor), and biological network analysis (in collaboration with Professors Tidor and Lauffenburger).

Many of our current projects involve biological applications. In the area of biomems design, we are currently investigating using fast integral equation methods combined with shape optimization techniques to design fluid channels for steering biological cells (with Professor Voldmann). We are developing fast solvers and matrix-implicit optimization algorithms for electrostatic analysis and optimization of biomolecules (with Professor Tidor). Finally, we are investigating model reduction techniques as applied to the nonlinear differential equation systems used to describe cellular transduction and regulation. In this last project, we hope to develop algorithmic techniques that will generate accurate models efficient enough to make possible the simulation of the cell collective behavior, such as wound healing and tumor growth.

For more information, see Computational Prototyping Group, CPG.

TRAINING GRANT

The Electrical Engineering and Computer Science Department at MIT is proud to have an NIH training grant that funds graduate students in the computational life sciences, broadly construed. This funding allows graduate students and faculty to explore new directions in life sciences research. The training program is an evolution, combination, and formalization of a successful training program between the Computer Science Department at MIT, the Biology Department at MIT, and the Whitehead Institute for Biomedical Research (WIBR), the Broad Institute. There are synergistic forces that have contributed to the creation of this new training program that originate both in Computer Science and in Biology at the Whitehead and Broad Institutes. The mission of our training program is to produce a new breed of interdisciplinary scientist who can create fundamentally new computational and mathematical approaches that enable significant forward progress on biological and health related problems. We have focused on this mission for over three years, and have successfully trained some of the most energetic and productive interdisciplinary young scientists that now have leadership positions in key universities and industry.

The success of our training program can be traced to three key factors. First, we have an extremely large, talented, and motivated pool of candidates that seek to work at the intersection of computer science, mathematics, and biology. Second, we have developed a strong co-mentorship and co-advising program for this candidate pool based upon long-standing and productive interdisciplinary collaborations between our laboratories. Finally, we have developed courses, held weekly interdisciplinary meetings, and run retreats that have catalyzed our students to produce excellent results and modeled a new kind of interdisciplinary science.

Graduates of our training program are now running interdisciplinary research programs at major universities as faculty (including Stanford, Duke, CMU, and Princeton) and in industry (Millennium Pharmaceuticals). We train graduate students who are highly technically accomplished to develop new algorithms, tools, and approaches for analyzing experimental biological data and expressing this analysis in the form of principled predictive models. The major research disciplines of this program include: 1) the development of new approaches and algorithms for the analysis of data from biological experiments, 2) approaches for the principled design of biological experiments based upon past data, 3) the construction of computational models that explain complex biological phenomenon, 4) and the development of approaches for interpreting clinical data relevant to human health and disease.