David Evans CS851: Biologically-Inspired Computing University of Virginia Computer Science Computing Inspired by Biology

15 April 2003CS 851 Spring Menu Computing with Biology Project Presentations Report on the NSF Advanced Computation Inspired by Biological Processes Workshop

15 April 2003CS 851 Spring Computing with DNA Leonard Adleman –2002 Turing Award Winner (for RSA) –Mathematical Consultant for Sneakers

15 April 2003CS 851 Spring DNA Sequence of nucleotides: adenine (A), guanine (G), cytosine (C), and thymine (T) Two strands, A must attach to T and G must attach to C A G T C

15 April 2003CS 851 Spring Hamiltonian Path Problem Input: a graph, start vertex and end vertex Output: either a path from start to end that touches each vertex in the graph exactly once, or false indicating no such path exists CHO RIC IAD BWI start: CHO end: BWI Hamiltonian Path is NP-Complete

15 April 2003CS 851 Spring Encoding The Graph Make up a two random 4-nucleotide sequences for each city: CHO: CHO 1 = ACTT CHO 2 = gcag RIC:RIC 1 = TCGGRIC 2 = actg IAD:IAD 1 = GGCTIAD 2 = atgt BWI:BWI 1 = GATCBWI 2 = tcca If there is a link between two cities (A  B), create a nucleotide sequence: A 2 B 1 CHO  RICgcagTCGG RIC  CHOactgACTT Based on Fred Hapgood’s notes on Adelman’s talk

15 April 2003CS 851 Spring Encoding The Problem Each city nucleotide sequence binds with its complement (A  T, G  C) : CHO: CHO 1 = ACTT CHO 2 = gcag CHO’: TGAA cgtc RIC: TCGGactg RIC’: AGCCtgac IAD: GGCTatgt IAD’ = CCGAtaca BWI: GATCtcca BWI’ = CTAGaggt Mix up all the link and complement DNA strands – they will bind to show a path!

15 April 2003CS 851 Spring Path Binding CHO RIC IAD BWI ACTTgcag TCGGactg GATCtcca GGCTatgt CHO’ TGAAcgtc gcagGGCT CHO  IAD IAD’ CCGAtaca atgtTCGG IAD  RIC RIC’ AGCCtgac BWI’ CTAGaggt actgGATC RIC  BWI

15 April 2003CS 851 Spring Getting the Solution Extract DNA strands starting with CHO and ending with BWI –Easy way is to remove all strands that do not start with CHO, and then remove all strands that do not end with BWI Measure remaining strands to find ones with the right weight (7 * 8 nucleotides) Read the sequence from one of these strands

15 April 2003CS 851 Spring Why don’t we solve NP- Complete problems this way? Speed: shaking up the DNA strands does operations per second ($400M supercomputer does ) Memory: we can store information in DNA at 1 bit per cubic nanometer How much DNA would you need? –Volume of DNA needed grows exponentially with input size –To solve ~45 vertices, you need ~20M gallons

15 April 2003CS 851 Spring DNA-Enhanced PC

15 April 2003CS 851 Spring Project Presentations Tell a story, don’t read a list.

15 April 2003CS 851 Spring All Good Talks Tell a Story Introduce characters (rabbit, fox) Describe an important problem (fox wants to eat rabbit) Relate events that resolve the problem (rabbit tells fox about thesis) A few examples (rabbit tells wolf,...) Draw a general conclusion that is supported by your story (thesis doesn’t matter, only advisor)

15 April 2003CS 851 Spring Introduction Introduce characters: motivate your work –Why the problem is interesting and important –Place your work in context: how is it different from what others have done Teaser for your results – why should we listen to the rest of the talk? –Don’t need a full outline, but let audience know enough so they want to stay –Unlike Rabbit story, suspense is not good

15 April 2003CS 851 Spring Guts Explain what you did –Don’t be comprehensive – big picture –Use pictures, 1-2 examples, etc. Convey one technical nugget –Show one neat concrete thing that came out of your work Analysis –Did your work solve the problem? –What are the important results of your work

15 April 2003CS 851 Spring Conclusion Summarize your project with one key point. Something your audience didn’t know or believe before your presentation If your audience remembers one thing from your talk, you have succeeded.

15 April 2003CS 851 Spring Can you do all this in 20 minutes? Advertisers pay $2.5M for 30 seconds during Superbowl – they must be pretty sure they can tell a compelling story in that time Seinfeld episode is 22 minutes long Make your points directly, avoid unnecessary details Organize your presentation

15 April 2003CS 851 Spring Some Specific Advice Average around 2 minutes per slide –No more than 15 slides total Your target audience is the other students in the class Use Pictures Use Humor (but only if its relevant) Don’t put more text than this on any of your slides!

15 April 2003CS 851 Spring Practice! Don’t just make up your slides, think about what you will say with them Without an audience In front of friends not familiar with your project With classmates

15 April 2003CS 851 Spring Project Presentations Sign up for a time slot today If your presentation and course contributions are enough to merit an “A” in the course already, you will not need to write a project report If not, you will have a second chance: project report

15 April 2003CS 851 Spring NSF Advanced Computation Inspired by Biological Processes Conference Arlington, VA 7-8 April 2003 Now, I’m going to give an unpracticed, list talk… 

15 April 2003CS 851 Spring NSF Bio-Inspired Computing Robust Cognitive Computation Development and Self-Organization Protection and Immunity –Stephanie Forrest, Computer Immunology Social Interaction –Eric Bonabeau, Science of “Prey”

15 April 2003CS 851 Spring Michale Fee, Bell Labs Zebra Finch Singing Electrodes can measure individual neurons Temporal firing of neurons corresponds to sections in song –Errors are localized Redundantly encoded – HVC neurons, 600ms song, each neuron is active only 6ms of the song –200 neurons scattered through HVC know each part of song

15 April 2003CS 851 Spring Alan Gelperin, Monell Chemical Senses Center Two types of olfactory systems: –Specialist (moth – male moth can only smell female moth, but very sensitive to it) –Generalist (dog) Olfaction is 5% of mouse brain, 0.2% of human brain No noticeable degradation when all of one bulb and 80% of the other is removed!

15 April 2003CS 851 Spring Theimo Krink University of Aarhus, Denmark Mass extinction in evolutionary algorithms End of Permian era (250M years): 96% of all species became extinct Experiment with mass random extinctions in evolutionary algorithms Better results (for toy problems) without extinction with extinction

15 April 2003CS 851 Spring Computer Immunology Stephanie Forrest

15 April 2003CS 851 Spring Traditional CS vs. Bio-Inspired CS Traditional Computer Science –Decomposability and Modularity –Explicit management of interactions and exceptions –Focus on performance and correctness Biology –Survivability and adaptability more important than performance and correctness

15 April 2003CS 851 Spring Biological Inspiration Biological systems are incredibly resilient Most humans survive ~80 years Before medical advances, most still would survive ~30 years Operate in a hostile, unpredictable environment No way to reboot, reinstall operating system, upgrade software, etc.

15 April 2003CS 851 Spring Immune Systems Lymphocytes recognize pathogens by binding. Proteins have distinctive shapes. Binding is approximate. Sometimes match wrong things (this is why organ transplants get rejected). An Overview of the Immune System. © 1997 Steven A Hofmeyr

15 April 2003CS 851 Spring Receptor Diversity Need to recognize all foreign intruders, but DNA can’t know about all possible intruders –10 16 intruder patterns, 10 6 self patterns Gene segments are randomly combined to form different receptors –Can make up to different receptors (usually have about 10 8 – at one time) How is that enough? –Matching is approximate –Create 100M new lymphocytes every day

15 April 2003CS 851 Spring How You Make Them

15 April 2003CS 851 Spring Better Not Kill Yourself Clonal Deletion

15 April 2003CS 851 Spring Affinity Maturation The ones that match intruders are produced in quantity B-cells in bone marrow – most effective cells reproduce more quickly An Overview of the Immune System. © 1997 Steven A Hofmeyr

15 April 2003CS 851 Spring Can computers do this? Programs identified by sequences of system calls Build a database of normal patterns (how?) Receptors recognize unusual patterns Enough unusual patterns is considered an intrusion

15 April 2003CS 851 Spring False Positives Sequence of system calls not recognized as a normal pattern generated by non- intrusive execution Do biological immune systems have false positives? Yes – that is what auto-immune diseases are: Multiple Sclerosis – motor nerve cells are non-self Grave’s Disease – thyroid gland is non-self Rhematoid Arthritis – connective tissue is non-self

15 April 2003CS 851 Spring Fatal Flaw of Intrusion Detection Might work okay if no one important is using it Will it work if an attacker knows about it and is deliberately constructing an attack to avoid detection? –Do biological viruses evolve to mimic host proteins?

Social Interaction Eric Bonabeau Swarm Intelligence (The Science of “Prey”)

15 April 2003CS 851 Spring Social Interaction Swarm Intelligence is a mindset, not a technology –bottom-up approach to controlling and optimizing distributed systems Optimization isn’t so interesting – think about doing new things

15 April 2003CS 851 Spring Self-Assembly: Game Game 1: pick a protector and aggressor, keep protector between yourself and aggressor Game 2: pick a protectee and aggressor, keep yourself between aggressor and protectee You Protector Aggressor Protectee You Aggressor

15 April 2003CS 851 Spring Charge Presentations Thursday and next week –Make them story talks not list talks!