ISI TALK - Dec, 19, 2001

Grand Problems of Biocomputing, 2002-2010 Dr. Pete Markiewicz Indiespace pete@indiespace.com

Basic outline - (full presentation with images at talk)

Summary

• Why biocomputing will expand relative to other computing realms
• Features of the “grand problems” (GP) of biocomputing
• 4 specific “grand problems” in genomics and proteomics
• Examples where new hardware & software solutions could prove useful

Background and interests

• PhD. in Biophysics & Theoretical Biology at U. Chicago (1984)
• Vaccine R & D at USAMRIID - "The Hot Zone" (1984 - 1988)
• UCLA Molecular & Cell Biology - protein structure/evolution (1988 - 1993)
• Internet-based technology & strategic marketing indiespace.com (1994 - 2001)

Strengths

• Seeing the “big picture”
• Associative connection across disciplines
• Knowledge of the 4 information processing systems on the planet (language, cells, immune systems, computers)

Features of "grand problems" (GP) in biology

• Too big for desktop computers
• Data naturally fits multiprocessor/networked computing paradigms
• Standard analysis is faltering
• May require exotic H/S for effective solution

'Follow the money'

• Dotcoms have cashed in their chips
• Long-term shift to life science in US
• Aging Boomer population will push health and “alternative” solutions
• 80-year shift point in economics

The 'forbidden top' and 'trivial bottom'

• Structural analysis of some biomolecules is well-understood (e.g. genome assembly)
• Expanding or improving networks doesn’t break new ground (e.g., wireless)
• Molecular “first principles” solution to 3-D structures (e.g. proteins) is beyond current computing technology

The "grand problem" middle ground

• Develop strategies for analyzing the uniquely biological features of the system.
• Look for problems that need different hardware/software approaches
• Look for problems which imply the "convergence" of biology and computing (e.g. two of the three information-processing systems on the planet) Cell computers, evolutionary programming

4 GPS for the 2002-2010 timeframe

• Evolvability versus engineerability/computability
• Structure determination via combined evolutionary/engineering analysis
• Models of complex cell regulatory systems
• Human-centric computing - a shift in computing metaphors

GP#1: Evolvability versus Engineerability

• Why don’t Nature & engineers take the same design path?
• Engineerability = modification via genetic engineering
• Evolvability = modification during evolutionary history
• Nature is not "clumsy" relative to human engineering

GP#1: Can we predict protein structure by combining engineering & evolution?

• Prediction of 3-D protein structure from sequence & “first principles” has hit a wall
• Evolution has already rebuilt the same proteins many times, from different parts
• Protein engineers mutate protein sequences to change structure and function

GP#2: How is the genome organized?

• 1000s of complete genomes will be available by 2010
• Tremendous sequence variability masks similarity of operation
• Determining exact location of structural genes & regulatory components essential for building gene expression models

GP#3: Modeling of gene regulatory networks

• Gene expression networks = modeling information processing in portions of a cell
• Gene expression involves hundreds of cell components with multiple, loose interactions
• Empirical “gene chip” & "protein chip" data provides window into time-dependent expression of some components (many hidden)

GP#4: The Unfinished Revolution - can we develop human-centric computing?

• Airport toilets smarter than Deep Blue
• Current computer systems require us to enter a virtual world in order to work
• Doctors, etc. are unwilling to go farther into cyberspace

Role of exotic hardware/software

• Most GPs are easily decomposed to parallel or networked solutions
• Hardware that directly embodies a biological process enjoys greater success (UPenn/Boahen’s vision chip)
• Evolutionary programming may duplicate information processing in biology

Specific biocomputing GP tracks

• Combine evolutionary and engineering data to predict protein structure
• Determine genome structure by evolutionary analysis
• Deduce gene expression networks from “gene chip” data
• Introduce a “human-centric” computing model into healthcare environments

GP Track I - Predict protein structure & function from primary sequence

• Optimization/”hill climbing” models based on chemistry have failed to fold proteins
• Mass crystalization studies (Eisenberg) will provide 1000s of 3-D protein structures by 2005
• Genome sequencing will provide an even larger (20-100,000) sequence database of proteins

GP Track I - Biocomputing strategy

• Align primary protein sequences according to evolutionary trees
• Align 3-D protein structures in “structure space”
• Align additional protein sequences without solved 3-D structures
• Combine aligned datasets into predictive models

GP Track II - Determine genome structure via evolutionary analysis

• Comparative anatomy helped to determine the function of organs in classical biology
• Molecular comparative anatomy may reveal the underlying structure of genome organization and regulatory systems
• 1000s of complete genome sequences will be available by 2010

GP Track II - Biocomputing strategy

• Create database for handling 1000s of gigabit-sized genome sequences
• Simultaneous cross analysis of gigabit genomes (super-clustering, dynamic pg.)
• Use sequence annotations in alignment via rule-based systems
• Output families of models rather than one optimal model

GP Track III -Determine gene expression from "gene chip/protein chip" data

• New “gene chips” reveal time-dependent expression of genes & their products
• Data potentially provides guide to structure of gene expression networks

GP Track III - Biocomputing strategy

• Develop databases to handle expression data for different organisms, conditions (data warehousing)
• Develop techniques for normalizing data for cross-comparison between events, species
• Use expression data to develop a “model space” of allowed networks fitting the data

GP Track IV - Introduce "human centric" computing model into healthcare environments

• Current attempts to implement wireless networks, databases in healthcare meets with resistance
• Systems need to fit the way healthcare professionals think, rather than bending their minds to the will of cyberspace

GP Track IV - Standard solutions are inappropriate

• Standard computing solutions ignore “biological” component
• Biocomputing needs different approach Doctor PDA/case notes example

GP Track IV - Biocomputing strategy

• Develop realtime OS systems that actively sense same environment as people (robots)
• Collect & recover information using associative models
• Present data as an advisor, rather than an oracle
• Convert existing literature to truly interactive systems (e.g. textbooks)

A few final thoughts

• Don't underestimate the importance of "dramatic" experiments (marketing value)
• Don't underestimate the importance of "style" and design (Japanese vs. US robots)