Biological sequence analysis

Terry Speed

Wald Lecture II,

August 8, 2001

The objects of our study DNA, RNA and proteins: macromolecules which are

unbranched polymers built up from smaller units.

DNA: units are the nucleotide residues A, C, G and T RNA: units are the nucleotide residues A, C, G and U Proteins: units are the amino acid residues A, C, D, E,

F, G, H, I, K, L, M, N, P, Q, R, S, T, V, W and Y.

To a considerable extent, the chemical properties of DNA, RNA and protein molecules are encoded in the linear sequence of these basic units: their primary structure.

The use of statistics to study linear sequences of biomolecular units

Can be descriptive, predictive or everything else in between…..almost business as usual.

Stochastic mechanisms should never be taken literally, but nevertheless can be amazingly useful.

Care is always needed: a model or method can break down at any time without notice.

Biological confirmation of predictions is almost always necessary.

The statistics of biological sequences can be global or local

Base composition of genomes:

E. coli: 25% A, 25% C, 25% G, 25% T

P. falciparum: 82%A+T

Translation initiation:

ATG is the near universal motif indicating the

start of translation in DNA coding sequence.

1 ZNF: Cys-Cys-His-His zinc finger DNA binding domain

From certainty to statistical models: a brief case study

Cys-Cys-His-His zinc finger DNA binding domain

Its characteristic motif has regular expression

C-x(2,4)-C-x(3)-[LIVMFYWC]-x(8)-H-x(3,5)-H

1ZNF: XYKCGLCERSFVEKSALSRHQRVHKNX

Regular expressions can be limiting, and position-specific

distributions came to represent the variability in motifs.

Cys-Cys-His-His profile: sequence logo form

A sequence logo is a scaled position-specific a.a.distribution. Scaling is by a measure of a position’s information content.

Representation of motifs: the next steps

Missing from the position-specific distribution

representation of motifs are good ways of dealing with:

Length distributions for insertions/deletions

Cross-position association of amino acids

Hidden Markov models help with the first. The second remains a hard unsolved problem.

Hidden Markov models

Processes {(St,Ot), t=1,…}, where St is the hidden

state and Ot the observation at time t, such that

pr(St | St-1,Ot-1,St-2 ,Ot-2 …) = pr(St | St-1)

pr(Ot | St-1,Ot-1,St-2 ,Ot-2 …) = pr(Ot | St, St-1)

The basics of HMMs were laid bare in a series of beautiful papers by L E Baum and colleagues around 1970, and their formulation has been used almost unchanged to this day.

Hidden Markov models:extensions

Many variants are now used. For example, the distribution of O may not depend on previous S but on previous O values,

pr(Ot | St , St-1 , Ot-1 ,.. ) = pr(Ot | St ), or

pr(Ot | St , St-1 , Ot-1 ,.. ) = pr(Ot | St , St-1 ,Ot-1) .

Most importantly for us, the times of S and O may be decoupled,

permitting the Observation corresponding to State time t to be a string whose length and composition depends on St (and possibly St-1 and part or all of the previous Observations). This is called a hidden semi-Markov or generalized hidden Markov model.

A simple HMM (Churchill, 1989)

. . . . . .

. . . . . .A T C A A G G C G A T

A 0.4C 0.1G 0.1T 0.4

A 0.05C 0.4G 0.5T 0.05

O.O1O.9

O.1

O.99

hidden states

observations

The algorithms

As the name suggests, with an HMM the series O = (O1,O2 ,O3 ,……., OT) is observed, while the states S = (S1 ,S2 ,S3 ,……., ST) are not.

There are elegant algorithms for calculating pr(O|), arg max pr(O|) in certain special cases, and arg maxS pr(S|O,).

Here are the parameters of the model, e.g. transition and observation probabilities.

Some current applications of HMMs to biology

mapping chromosomes

aligning biological sequences

predicting sequence structure

inferring evolutionary relationships

finding genes in DNA sequence

Some early applications of HMMs

finance, but we never saw them

speech recognition

modelling ion channels

In the mid-late 1980s HMMs entered genetics and molecular biology, and they are now firmly entrenched.

2TMA: Tropomyosin

HMMs representing coiled-coil domains

Coiled-coil domains, schematically

dimeric parallel helices, heptad repeats, knobs-into-holes

(hydrophobic)

Designing the HMM, I

Designingthe HMM, 2

Designing the HMM, 3

HMM: decoding

WGP ARQLNES VKDINKM LER HPBBB CCCCCCC CCCCCCC CCC BB000 abcdefg abcdefg abc 0000c defgabc defgabc def g0

Sequence Labels Path1 Path2

VITERBI decoding: of all possible state-paths, we determine the maximum probability one given the amino acid sequence O;

POSTERIOR decoding: at each position, we determine the state with the highest probability given O.

Issue: how to measure the strength of a potential CC domain, and how should this depend on the length of a domain?

CC-PROBABILITY PROFILE

Fusion protein of simian parainfluenza virus 5

Assessing performance: terms

TP true positive: a predicted fragment that overlaps the annotated fragment (aa in the

annotated region) FP false positive: a predicted fragment does not

overlap any annotated fragment (aa not in the annotated region)

LS learning set of sequences

NTS negative test set; sequences with no CCD used for estimating FP

PTS positive test set; used for estimating TPMuch care/effort required to create these sets

Assessing performance: study design Study the variability of performance under variation of the sequences used for determining

the model parameters.

Compare methods using the same set of aa-frequencies / emission probabilities. Use the same set of domains for learning and testing instead of testing on different protein

families.

Choose a number of FP-rates and calculate the corresponding TP-rates (ROC curve).

PTS subdivided 150 times at random (stratified)into 2/3 for learning and the rest for testing.

ANALYSIS OF RESULTS

LEARNING PHASE => PARAMETERS

TESTING ON NTS => THRESHOLDS

TESTING ON PTS => TP-VALUES

150 X

Assessing performance: summaries

TP-rate at given FP-rate: per family / per length-class

TP- and FP-rates for aa’s: per family / per length-class

Accuracy: at the borders and of length prediction

Finding genes in DNA sequence

This is one of the most challenging and interesting problems in computational biology at the moment. With so many genomes being sequenced so rapidly, it remains important to begin by identifying genes computationally.

What is a (protein-coding) gene?

Protein

mRNA

DNA

transcription

translation

CCTGAGCCAACTATTGATGAA

PEPTIDE

CCUGAGCCAACUAUUGAUGAA

What is a gene, ctd?

In general the transcribed sequence is longer than the translated portion: parts called introns (intervening sequence) are removed, leaving exons (expressed sequence), and yet other regions remain untranslated. The translated sequence comes in triples called codons, beginning and ending with a unique start (ATG) and one of three stop (TAA, TAG, TGA) codons.

There are also characteristic intron-exon boundaries called splice donor and acceptor sites, and a variety of other motifs: promoters, transcription start sites, polyA sites,branching sites, and so on.

All of the foregoing have statistical characterizations.

In more detail(color ~state)

Some facts about human genes

Comprise about 3% of the genome

Average gene length: ~ 8,000 bp

Average of 5-6 exons/gene

Average exon length: ~200 bp

Average intron length: ~2,000 bp

~8% genes have a single exon

Some exons can be as small as 1 or 3 bp.

HUMFMR1S is not atypical: 17 exons 40-60 bp long, comprising 3% of a 67,000 bp gene

The idea behind a GHMM genefinder

States represent standard gene features: intergenic region, exon, intron, perhaps more (promotor, 5’UTR, 3’UTR, Poly-A,..).

Observations embody state-dependent base composition, dependence, and signal features.

In a GHMM, duration must be included as well. Finally, reading frames and both strands must

be dealt with.

Half a model for a genefinder

Splice sites can be included in the exons

Beyond position-specific distributions

The bases in splice sites exhibit dependence, and not simply of the nearest neighbor kind.

High-order (non-stationary) Markov models would be one option, but the number of parameters in

relation to the amount of data rules them out. The class of variable length Markov models (VLMMs) deriving from early research by Rissanen prove to be valuable in this context. However, there is

likely to be room for more research here.

E0 E1 E2

E

poly-A

3'UTR5'UTR

tEi

Es

I0 I 1 I 2

intergenicregion

Forward (+) strand

Reverse (-) strand

Forward (+) strand

Reverse (-) strand

promoter

62001 AGGACAGGTA CGGCTGTCAT CACTTAGACC TCACCCTGTG GAGCCACACC

62051 CTAGGGTTGG CCAATCTACT CCCAGGAGCA GGGAGGGCAG GAGCCAGGGC

62101 TGGGCATAAA AGTCAGGGCA GAGCCATCTA TTGCTTACAT TTGCTTCTGA

62151 CACAACTGTG TTCACTAGCA ACCTCAAACA GACACCATGG TGCACCTGAC

62201 TCCTGAGGAG AAGTCTGCCG TTACTGCCCT GTGGGGCAAG GTGAACGTGG

62251 ATGAAGTTGG TGGTGAGGCC CTGGGCAGGT TGGTATCAAG GTTACAAGAC

62301 AGGTTTAAGG AGACCAATAG AAACTGGGCA TGTGGAGACA GAGAAGACTC

62351 TTGGGTTTCT GATAGGCACT GACTCTCTCT GCCTATTGGT CTATTTTCCC

62401 ACCCTTAGGC TGCTGGTGGT CTACCCTTGG ACCCAGAGGT TCTTTGAGTC

62451 CTTTGGGGAT CTGTCCACTC CTGATGCTGT TATGGGCAAC CCTAAGGTGA

62501 AGGCTCATGG CAAGAAAGTG CTCGGTGCCT TTAGTGATGG CCTGGCTCAC

62551 CTGGACAACC TCAAGGGCAC CTTTGCCACA CTGAGTGAGC TGCACTGTGA

62601 CAAGCTGCAC GTGGATCCTG AGAACTTCAG GGTGAGTCTA TGGGACCCTT

62651 GATGTTTTCT TTCCCCTTCT TTTCTATGGT TAAGTTCATG TCATAGGAAG

62701 GGGAGAAGTA ACAGGGTACA GTTTAGAATG GGAAACAGAC GAATGATTGC

62751 ATCAGTGTGG AAGTCTCAGG ATCGTTTTAG TTTCTTTTAT TTGCTGTTCA

62801 TAACAATTGT TTTCTTTTGT TTAATTCTTG CTTTCTTTTT TTTTCTTCTC

62851 CGCAATTTTT ACTATTATAC TTAATGCCTT AACATTGTGT ATAACAAAAG

62901 GAAATATCTC TGAGATACAT TAAGTAACTT AAAAAAAAAC TTTACACAGT

62951 CTGCCTAGTA CATTACTATT TGGAATATAT GTGTGCTTAT TTGCATATTC

63001 ATAATCTCCC TACTTTATTT TCTTTTATTT TTAATTGATA CATAATCATT

63051 ATACATATTT ATGGGTTAAA GTGTAATGTT TTAATATGTG TACACATATT

63101 GACCAAATCA GGGTAATTTT GCATTTGTAA TTTTAAAAAA TGCTTTCTTC

63151 TTTTAATATA CTTTTTTGTT TATCTTATTT CTAATACTTT CCCTAATCTC

63201 TTTCTTTCAG GGCAATAATG ATACAATGTA TCATGCCTCT TTGCACCATT

63251 CTAAAGAATA ACAGTGATAA TTTCTGGGTT AAGGCAATAG CAATATTTCT

63301 GCATATAAAT ATTTCTGCAT ATAAATTGTA ACTGATGTAA GAGGTTTCAT

63351 ATTGCTAATA GCAGCTACAA TCCAGCTACC ATTCTGCTTT TATTTTATGG

63401 TTGGGATAAG GCTGGATTAT TCTGAGTCCA AGCTAGGCCC TTTTGCTAAT

63451 CATGTTCATA CCTCTTATCT TCCTCCCACA GCTCCTGGGC AACGTGCTGG

63501 TCTGTGTGCT GGCCCATCAC TTTGGCAAAG AATTCACCCC ACCAGTGCAG

63551 GCTGCCTATC AGAAAGTGGT GGCTGGTGTG GCTAATGCCC TGGCCCACAA

63601 GTATCACTAA GCTCGCTTTC TTGCTGTCCA ATTTCTATTA AAGGTTCCTT

63651 TGTTCCCTAA GTCCAACTAC TAAACTGGGG GATATTATGA AGGGCCTTGA

63701 GCATCTGGAT TCTGCCTAAT AAAAAACATT TATTTTCATT GCAATGATGT

GENSCAN (Burge & Karlin)

Remark

In general the problem of identifying (annotating) human genes is considerably harder than ß-globin might suggest.

The human factor VIII gene (whose mutations cause hemophilia A) is spread over ~186,000 bp. It consists of 26 exons ranging in size from 69 to 3,106 bp, and its 25 introns range in size from 207 to 32,400 bp. The complete gene is thus ~9 kb of exon and ~177 kb of intron.

The biggest human gene yet is for dystrophin. It has

> 30 exons and is spread over 2.4 million bp.

Challenges in the analysis of sequence data

Understanding the biology well enough to begin.

Designing HMM architecture, e.g. in Marcoil for coiled-coils.

Modelling the parts, e.g. VLMMs for splice sites.

Coding = software engineering, is the hardest and most important task of all: making it all work.

Obtaining good data sets for use in careful evaluation and comparison with competing algorithms; designing the studies.

Opportunities for methodological research.

Topics not mentioned include

Molecular evolution, including phylogenetic inference (building trees from aligned sequence data)

Sequence alignment (pairwise, multiple), including use of Gibbs sampler

Stochastic context-free grammar models and the analysis of RNA sequence data.

Acknowledgements

Mauro Delorenzi (WEHI)

Simon Cawley (Affymetrix)

Tony Wirth (CS, Princeton)

Lior Pachter (Math, UCB)

Marina Alexandersson (Stat, UCB)

ReferencesBiological Sequence AnalysisR Durbin, S Eddy, A Krogh and G MitchisonCambridge University Press, 1998.

Bioinformatics The machine learning approach

P Baldi and S BrunakThe MIT Press, 1998

Post-Genome InformaticsM KanehisaOxford University Press, 2000