master’s course bioinformatics data analysis and tools
DESCRIPTION
C. E. N. T. E. R. F. O. R. I. N. T. E. G. R. A. T. I. V. E. B. I. O. I. N. F. O. R. M. A. T. I. C. S. V. U. Master’s course Bioinformatics Data Analysis and Tools. Lecture 1: Introduction Centre for Integrative Bioinformatics FEW/FALW [email protected]. - PowerPoint PPT PresentationTRANSCRIPT
Master’s course
Bioinformatics Data Analysis and Tools
Lecture 1: Introduction
Centre for Integrative BioinformaticsFEW/FALW
CENTR
FORINTEGRATIVE
BIOINFORMATICSVU
E
Course objectives
• There are two extremes in bioinformatics work– Tool users (biologists): know how to press the
buttons and know the biology but have no clue what happens inside the program
– Tool shapers (informaticians): know the algorithms and how the tool works but have no clue about the biology
Both extremes can be dangerous at times, need a breed that can do both
Course objectives• How do you become a good bioinformatics
problem solver?– You need to know basic analysis and data mining
modes– You need to know some important backgrounds of
analysis and prediction techniques (e.g. statistical thermodynamics)
– You need to have knowledge of what has been done and what can be done (and what not)
• Is this enough to become a creative tool developer?– Need to like doing it– Experience helps
Course objectivesThe most important thing in tools creating (and
science in general)
• Be able to ask the proper question!– Should address a real problem– Should be targeted– Should be solvable
• Bioinformatics challenge: from Genomics to Systems Biology– Bottom up: start at the components, assemble and learn
the system– Top down: observe the system behaviour, model and
learn the details– What about bottom down or top up questions?
Contents (tentative dates)Date Lecture Title Lecturer
1 [wk 19] 01/04/08 Introduction Jaap Heringa
2 [wk 19] 03/04/08 Microarray data analysis Jaap Heringa
3 [wk 20] 08/04/08 Machine learning Jaap heringa
4 [wk 21] 10/04/08 Clustering algorithms Bart van Houte
5 [wk 21] 15/04/08 Feature Selection Bart van Houte
6[wk 23] 17/04/08 Molecular Simulation & Sampling Techniques Anton Feenstra
7[wk 23] 22/04/08 Introduction to Statistical Thermodynamics I Anton Feenstra
8[wk 24] 24/04/08 Introduction to Statistical Thermodynamics II Anton Feenstra
9[wk 24] 06/05/08 Databases and parsing Sandra Smit
10[wk 24] 08/05/08 Semantic Web and Ontologies Frank van Harmelen
11[wk 25] 13/05/08 Parallelisation& Grid Computing Thilo Kielmann
12[wk 25] 15/05/08 Application area I: Protein Domain Prediction Jaap Heringa
13[wk 25] 20/05/08 Application Area II: Repeats Detection Jaap Heringa
At the end of this course…
• You will have seen a couple of algorithmic examples• You will have got an idea about methods used in the
field• You will have a firm basis of the physics and
thermodynamics behind a lot of processes and methods• You will have learned about state-of-the-art
computational issues, such as Semantic Web and HTP Computing
• You will have an idea of and some experience as to what it takes to shape a bioinformatics tool
Bioinformatics
“Studying informatic processes in biological systems”
(Hogeweg)
Applying algorithms and mathematical formalisms tobiology (genomics)
“Information technology applied to the management and analysis of biological data” (Attwood and Parry-Smith)
This course
• General theory of crucial algorithms (GA, NN, HMM, SVM, etc..)
• Method examples• Research projects within own group
– Repeats– Domain boundary prediction
• Physical basis of biological processes and of (stochastic) tools
BioinformaticsLarge - external(integrative) Science Human
Planetary Science Cultural Anthropology
Population Biology Sociology Sociobiology Psychology Systems Biology Biology Medicine
Molecular Biology Chemistry Physics
Small – internal (individual)
Bioinformatics
Genomic Data Sources
• DNA/protein sequence
• Expression (microarray)
• Proteome (xray, NMR,
mass spectrometry)
• PPI
• Metabolome
• Physiome (spatial,
temporal)
Integrative bioinformatics
Protein structural data explosion
Protein Data Bank (PDB): 14500 Structures (6 March 2001)10900 x-ray crystallography, 1810 NMR, 278 theoretical models, others...
MathematicsStatistics
Computer ScienceInformatics
BiologyMolecular biology
Medicine
Chemistry
Physics
Bioinformatics
Bioinformatics inspiration and cross-fertilisation
Joint international programming initiatives
• Bioperlhttp://www.bioperl.org/wiki/Main_Pagehttp://bioperl.org/wiki/How_Perl_saved_human_genome
• Biopythonhttp://www.biopython.org/
• BioTclhttp://wiki.tcl.tk/12367
• BioJavawww.biojava.org/wiki/Main_Page
Algorithms in bioinformatics• string algorithms• dynamic programming• machine learning (NN, k-NN, SVM, GA, ..)• Markov chain models• hidden Markov models• Markov Chain Monte Carlo (MCMC) algorithms• stochastic context free grammars• EM algorithms• Gibbs sampling• clustering• tree algorithms (suffix trees)• graph algorithms• text analysis• hybrid/combinatorial techniques and more…
Some techniques can be reapplied to many different problems, e.g. clustering, NN, etc.
Algorithms in bioinformatics
• Approaches in methods can be reapplied, e.g. kowledge-based (inverse) prediction
• Different approaches can be combined, e.g. consensus prediction, metaservers
DBT
hits
PSSM
Q
Discarded sequences
Run query sequence against
database
Run PSSM against database
PSI-BLAST iterationPSI-BLAST iteration
Fold recognition by threading:THREADER and GenTHREADER
Query sequence
Compatibility scores
Fold 1
Fold 2
Fold 3
Fold N
Polutant recognition by microarray mapping:
Compatibility scores
Cond. 1
Cond. 2
Cond. 3
Cond. N
Contaminant 1
Contaminant 2
Contaminant 3
Contaminant N
Query array
Protein-protein interaction prediction
• Two extreme approaches, phlogenetic prediction and Molecular Dynamics (MD) simulation
• Mesoscopic modelling
• Soft-core Molecular Dynamics (MD)– Fuzzy residues
– Fuzzy (surface)
locations
ENFIN WP5 - BioRange (Anton Feenstra)
• Protein-protein interaction prediction
• Mesoscopic modelling
• Soft-core Molecular Dynamics (MD)– Fuzzy residues– Fuzzy (surface) locations
Where are important new questions?
New neighbouring disciplines• Translational Medicine
A branch of medical research that attempts to more directly connect basic research to patient care. Translational medicine is growing in importance in the healthcare industry, and is a term whose precise definition is in flux. In particular, in drug discovery and development, translational medicine typically refers to the "translation" of basic research into real therapies for real patients. The emphasis is on the linkage between the laboratory and the patient's bedside, without a real disconnect. This is often called the "bench to bedside" definition.
• Computational Systems BiologyComputational systems biology aims to develop and use efficient algorithms, data structures and communication tools to orchestrate the integration of large quantities of biological data with the goal of modeling dynamic characteristics of a biological system. Modeled quantities may include steady-state metabolic flux or the time-dependent response of signaling networks. Algorithmic methods used include related topics such as optimization, network analysis, graph theory, linear programming, grid computing, flux balance analysis, sensitivity analysis, dynamic modeling, and others.
• Neuro-informatics Neuroinformatics combines neuroscience and informatics research to develop and apply the advanced tools and
approaches that are essential for major advances in understanding the structure and function of the brain
Translational Medicine
• “From bench to bed side”
• Genomics data to patient data
• Integration
Natural progression of a gene
TERTIARY STRUCTURE (fold)TERTIARY STRUCTURE (fold)
Genome
Expressome
Proteome
Metabolome
Functional GenomicsFunctional GenomicsFrom gene to functionFrom gene to function
Systems Biologyis the study of the interactions between the components of a biological system, and how these interactions give rise to the function and behaviour of that system (for example, the enzymes and metabolites in a metabolic pathway). The aim is to quantitatively understand the system and to be able to predict the system’s time processes
• the interactions are nonlinear• the interactions give rise to emergent properties,
i.e. properties that cannot be explained by the components in the system
Systems Biologyunderstanding is often achieved through modeling and simulation of the system’s components and interactions.
Many times, the ‘four Ms’ cycle is adopted:
Measuring
Mining
Modeling
Manipulating
Neuroinformatics
• Understanding the human nervous system is one of the greatest challenges of 21st century science.
• Its abilities dwarf any man-made system - perception, decision-making, cognition and reasoning.
• Neuroinformatics spans many scientific disciplines - from molecular biology to anthropology.
Neuroinformatics• Main research question: How does the brain and
nervous system work?• Main research activity: gathering neuroscience data,
knowledge and developing computational models and analytical tools for the integration and analysis of experimental data, leading to improvements in existing theories about the nervous system and brain.
• Results for the clinic: Neuroinformatics provides tools, databases, models, networks technologies and models for clinical and research purposes in the neuroscience community and related fields.
Bioinformatics algorithms
For problems such as alignenment, secondary/tertiary structure prediction, phylogenetic tree determination, etc.
Algorithmic main components:
• Search function
• Scoring function
Bioinformatics algorithms
• Search function– Search space can be large– High time complexity of algorithms, or even
NP-complete/NP-hard problems
• Scoring function– Also called the Objective Function– Often the most important
Pair-wise alignmentComplexity of the problem
Combinatorial explosion- 1 gap in 1 sequence: n+1 possibilities- 2 gaps in 1 sequence: (n+1)n - 3 gaps in 1 sequence: (n+1)n(n-1), etc.
2n (2n)! 22n
= ~ n (n!)2
n
2 sequences of 300 a.a.: ~1088 alignments 2 sequences of 1000 a.a.: ~10600 alignments!
T D W V T A L KT D W L - - I K
Levinthal’s paradox (1969)
•Denatured protein refolds in ~ 0.1 – 1000 seconds
•Protein with e.g. 100 amino acids each with 2 torsions ( en )
Each can assume 3 conformations (1 trans, 2 gauche)3100x2 1095 possible conformations!
•Or: 100 amino acids with 3 possibilities in Ramachandran plot (coil): 3100 » 1047 conformations
•If the protein can visit one conformation in one ps (10-12s) exhaustive search costs 1047 x 10-12 s = 1035 s 1027 years!
(the lifetime of the universe 1010 years…)
Phylogenetic tree combinatorial explosion
Number of unrooted trees =
!32
!523
n
nn
Number of rooted trees =
!22
!322
n
nn
Combinatoric explosion
# sequences # unrooted # rooted trees trees
2 1 13 1 34 3 155 15 1056 105 9457 945 10,3958 10,395 135,1359 135,135 2,027,02510 2,027,025 34,459,425
A recap on Scoring and BenchmarkingQUERY
DATABASE
True Positive
True Negative
True Positive
False Positive
True Negative False Negative
T
POSITIVES
NEGATIVES
Evaluating multiple alignments (MSAs)Evaluating multiple alignments (MSAs)• Conflicting standards of truth
– evolution
– structure
– function
• With orphan sequences no additional information• Benchmarks depending on reference alignments• Quality issue of available reference alignment databases• Different ways to quantify agreement with reference
alignment (sum-of-pairs, column score)• “Charlie Chaplin” problem
Charlie Chaplin once joined a Charlie-Chaplin competition in disguise and became third. What does this tell you about the jury’s ‘objective function’ ?
Evaluation measuresQuery Reference
Column score‘strict’ measure
Sum-of-Pairs scoremore lenient measure
What fraction of the matched amino acid pairs (or alignment columns) in the reference MSA are recreated in the query MSA?
Scoring a single MSA with the Sum-of-pairs (SP) score
Sum-of-Pairs score
• Calculate the sum of all pairwise alignment scores
• This is equivalent to taking the sum of all matched a.a. pairs
• This can be done using gap penalties or not
Good alignments should have a high SP score, but it is not always the case that the true biological alignment has the highest score.
BAliBASE benchmark alignmentsBAliBASE benchmark alignmentsThompson et al. (1999) NAR 27, 2682.Thompson et al. (1999) NAR 27, 2682.
88 categories: categories:• cat. 1 - equidistantcat. 1 - equidistant
• cat. 2 - orphan sequencecat. 2 - orphan sequence
• cat. 3 - 2 distant groupscat. 3 - 2 distant groups
• cat. 4 – long overhangscat. 4 – long overhangs
• cat. 5 - long insertions/deletionscat. 5 - long insertions/deletions
• cat. 6 – repeatscat. 6 – repeats
• cat. 7 – transmembrane proteinscat. 7 – transmembrane proteins
• cat. 8 – circular permutationscat. 8 – circular permutations
Evaluating multiple alignmentsEvaluating multiple alignments
You can score a single MSA using the sum of all matched amino acid pairs score. This is also referred to as the Sum-of-Pairs (SP) score.. (a bit confusing with the SP score for comparing a query alignment with a reference alignment )
Evaluating multiple alignmentsEvaluating multiple alignments
SP
BAliBASE alignment nseq * len
Many test alignments have a higher SP score than the reference alignment (“Charlie Chaplin problem’)
Evaluating multiple alignmentsEvaluating multiple alignments
Many test alignments have a higher SP score than the reference alignment (“Charlie Chaplin problem’)
Comparing T-coffee with other methods
Column scores are used here
BAliBASE benchmark alignments
If you are a better program on average, this does not mean you win in all cases…
How do you know what is the situation in your case? Even with a better program you can be unlucky..
Example from course DPSFAP: inverse folding
Top score structure 20 a.a. fragments in the high specificity regions -- Sequence: 3icb (residues 31–50)Protein Starting position Score Cr.m.s.d. Secondary structure (DSSP)
to native (A° )
3icb 31 –7.36 0.00 HHHHH TTTSSSSS HHHHH
1bbk B 32 –6.18 5.65 GGT SSS TT EE S E
1ezm 254 –5.93 4.61 HHHHT TT HHHHHHHHH
8cat A 73 –5.84 8.68 SEEEEEEEEEE S TTT
3enl 196 –5.84 3.82 HHHHHH GGGG B TTS B
1tie 59 –5.75 6.17 EESS SS TT EEEEES
3gap A 97 –5.73 3.11 EEHHHHHHHTTT TTTHHHH
1tfd 71 –5.59 6.50 EEEEEEE S SSS S E
1gsr A 159 –5.54 2.93 HHHHH TTTTTT HHHHHHH
1apb 149 –5.53 4.14 HHHHHHHHHHHHTT GGGE
Random 5.88 A°
The native structure is on top
Top-scoring structural 20 a.a. fragments in regions where the native state does not have lowest scores but the CRMSDs are low -- Sequence: 3icb (residues 36–55) Protein Starting position Score Cr.m.s.d. Secondary structure (DSSP)
to native (A° )
1mba 75 –9.54 3.16 HHHHTT HHHHHHHHHHHHH
1mbc 72 –8.59 3.84 HHHHTTT TTTHHHHHHHHH
3gap A 102 –8.43 3.54 HHHHTTT TTTHHHHHHHHH
1ezm 186 –7.83 5.44 ETTTTBSSS SEESSSGGG
1hmd A 67 –7.47 4.76 TTHHHHHHHHHHHHHHHHT
1sdh A 37 –7.42 4.65 HHHHHHH GGGGGGGGGG
2ccy A 36 –7.34 4.38 TTHHHHHHHHHHHHHHGGG
1ama 298 –7.11 2.67 HHHHHHSHHHHHHHHHHHHH
3icb 36 –7.08 0.00 TTTSSSSS HHHHHHHH S
1pbx A 30 –7.06 4.79 HHHHHHH GGGGGGSTTSS Random RMSD: 5.79 A°
The native structure is not on top
Inaccurate scoring functions• Since most scoring functions do not rank solutions
properly, often ensembles of predictions are taken– For example, take the top 10 (or top 5%) predicted
structures, is the true structure among those?– In alignment: next to optimal (highest scoring)
alignment, the ensemble of supoptimal alignments is taken
• Biology is also capricious, e.g. the native protein structure does not always have the lowest internal energy
• Integrate data sources
• Integrate methods
• Integrate data through method integration (biological model)
Integrative bioinformatics
Data
Algorithm
BiologicalInterpretation
(model)
tool
Integrative bioinformaticsData integration
Integrative bioinformaticsData integration
Data 1 Data 2 Data 3
Integrative bioinformaticsData integration
Data 1
Algorithm 1
BiologicalInterpretation
(model) 1
tool
Algorithm 2
BiologicalInterpretation
(model) 2
Algorithm 3
BiologicalInterpretation
(model) 3
Data 2 Data 3
“Nothing in Biology makes sense except in the light of evolution” (Theodosius Dobzhansky (1900-1975))
“Nothing in Bioinformatics makes sense except in the light of Biology”
Bioinformatics