1. Gene prediction

2. Gene Prediction: Computational Challenge aatgcatgcggctatgctaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggctatg ctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgggatccgatgacaatgcatgc ggctatgctaatgaatggtcttgggatttaccttggaatgctaagctgggatccgatgacaatgcatgcgg ctatgctaatgaatggtcttgggatttaccttggaatatgctaatgcatgcggctatgctaagctgggatccg atgacaatgcatgcggctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctg cggctatgctaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggctatgctaatgcat gcggctatgcaagctgggatcctgcggctatgctaatgaatggtcttgggatttaccttggaatgctaagct gggatccgatgacaatgcatgcggctatgctaatgaatggtcttgggatttaccttggaatatgctaatgca tgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggcta tgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggctatgctaatgcatgcg gctatgctaagctcatgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatga caatgcatgcggctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggc tatgctaatgcatgcggctatgctaagctcggctatgctaatgaatggtcttgggatttaccttggaatgcta agctgggatccgatgacaatgcatgcggctatgctaatgaatggtcttgggatttaccttggaatatgctaa tgcatgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcg gctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggctatgctaatgca tgcggctatgctaagctcatgcgg

3. Gene Prediction: Computational Challenge aatgcatgcggctatgctaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggctatg ctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgggatccgatgacaatgcatgc ggctatgctaatgaatggtcttgggatttaccttggaatgctaagctgggatccgatgacaatgcatgcgg ctatgctaatgaatggtcttgggatttaccttggaatatgctaatgcatgcggctatgctaagctgggatccg atgacaatgcatgcggctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctg cggctatgctaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggctatgctaatgcat gcggctatgcaagctgggatcctgcggctatgctaatgaatggtcttgggatttaccttggaatgctaagct gggatccgatgacaatgcatgcggctatgctaatgaatggtcttgggatttaccttggaatatgctaatgca tgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggcta tgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggctatgctaatgcatgcg gctatgctaagctcatgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatga caatgcatgcggctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggc tatgctaatgcatgcggctatgctaagctcggctatgctaatgaatggtcttgggatttaccttggaatgcta agctgggatccgatgacaatgcatgcggctatgctaatgaatggtcttgggatttaccttggaatatgctaa tgcatgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcg gctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggctatgctaatgca tgcggctatgctaagctcatgcgg

4. Gene Prediction: Computational Challenge aatgcatgcggctatgctaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggctatg ctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgggatccgatgacaatgcatgc ggctatgctaatgaatggtcttgggatttaccttggaatgctaagctgggatccgatgacaatgcatgcgg ctatgctaatgaatggtcttgggatttaccttggaatatgctaatgcatgcggctatgctaagctgggatccg atgacaatgcatgcggctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctg cggctatgctaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggctatgctaatgcat gcggctatgcaagctgggatcctgcggctatgctaatgaatggtcttgggatttaccttggaatgctaagct gggatccgatgacaatgcatgcggctatgctaatgaatggtcttgggatttaccttggaatatgctaatgca tgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcggcta tgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggctatgctaatgcatgcg gctatgctaagctcatgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatga caatgcatgcggctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggc tatgctaatgcatgcggctatgctaagctcggctatgctaatgaatggtcttgggatttaccttggaatgcta agctgggatccgatgacaatgcatgcggctatgctaatgaatggtcttgggatttaccttggaatatgctaa tgcatgcggctatgctaagctgggaatgcatgcggctatgctaagctgggatccgatgacaatgcatgcg gctatgctaatgcatgcggctatgcaagctgggatccgatgactatgctaagctgcggctatgctaatgca tgcggctatgctaagctcatgcgg Gene!

5. Newspaper written in unknown language –Certain pages contain encoded message, say 99 letters on page 7, 30 on page 12 and 63 on page 15. How do you recognize the message? You could probably distinguish between the ads and the story (ads contain the “$” sign often) Statistics-based approach to Gene Prediction tries to make similar distinctions between exons and introns. Gene Prediction Analogy

6. Noting the differing frequencies of symbols (e.g. ‘%’, ‘.’, ‘-’) and numerical symbols could you distinguish between a story and the stock report in a foreign newspaper? Statistical Approach: Metaphor in Unknown Language

7. Statistical: coding segments (exons) have typical sequences on either end and use different subwords than non-coding segments (introns). Similarity-based: many human genes are similar to genes in mice, chicken, or even bacteria. Therefore, already known mouse, chicken, and bacterial genes may help to find human genes. Two Approaches to Gene Prediction

8. If you could compare the day’s news in English, side-by-side to the same news in a foreign language, some similarities may become apparent Similarity-Based Approach: Metaphor in Different Languages

9. Annotation of Genomic Sequence Given the sequence of an organism’s genome, we would like to be able to identify: –Genes –Exon boundaries & splice sites –Beginning and end of translation –Alternative splicings –Regulatory elements (e.g. promoters) The only certain way to do this is experimentally, but it is time consuming and expensive. Computational methods can achieve reasonable accuracy quickly, and help direct experimental approaches. primary goals secondary goals

10. Prokaryotic Gene Structure Promoter CDS Terminator transcription Genomic DNA mRNA  Most bacterial promoters contain the Shine-Delgarno signal, at about -10 that has the consensus sequence: 5'-TATAAT-3'.  The terminator: a signal at the end of the coding sequence that terminates the transcription of RNA  The coding sequence is composed of nucleotide triplets. Each triplet codes for an amino acid. The AAs are the building blocks of proteins.

11. Pieces of a (Eukaryotic) Gene (on the genome) 5’ 3’ 5’ ~ 1-100 Mbp 5’ 3’ 5’ … … … … ~ 1-1000 kbp exons (cds & utr) / introns (~ 10 2 -10 3 bp) (~ 10 2 -10 5 bp) Polyadenylation site promoter (~10 3 bp) enhancers (~10 1 -10 2 bp) other regulatory sequences (~ 10 1 -10 2 bp)

12. What is Computational Gene Finding? Given an uncharacterized DNA sequence, find out: –Which region codes for a protein? –Which DNA strand is used to encode the gene? –Which reading frame is used in that strand? –Where does the gene starts and ends? –Where are the exon-intron boundaries in eukaryotes? –(optionally) Where are the regulatory sequences for that gene?

13. Prokaryotic Vs. Eukaryotic Gene Finding Prokaryotes: small genomes 0.5 – 10·10 6 bp high coding density (>90%) no introns –Gene identification relatively easy, with success rate ~ 99% Problems: overlapping ORFs short genes finding TSS and promoters Eukaryotes: large genomes 10 7 – 10 10 bp low coding density (<50%) intron/exon structure –Gene identification a complex problem, gene level accuracy ~50% Problems: many

14. What is it about genes that we can measure (and model)? Most of our knowledge is biased towards protein-coding characteristics –ORF (Open Reading Frame): a sequence defined by in- frame AUG and stop codon, which in turn defines a putative amino acid sequence. –Codon Usage: most frequently measured by CAI (Codon Adaptation Index) Other phenomena –Nucleotide frequencies and correlations: value and structure –Functional sites: splice sites, promoters, UTRs, polyadenylation sites

15. General Things to Remember about (Protein-coding) Gene Prediction Software It is, in general, organism-specific It works best on genes that are reasonably similar to something seen previously It finds protein coding regions far better than non- coding regions In the absence of external (direct) information, alternative forms will not be identified It is imperfect! (It’s biology, after all…)

16. Gene Finding: Different Approaches Similarity-based methods (extrinsic) - use similarity to annotated sequences : –proteins –cDNAs –ESTs Comparative genomics - Aligning genomic sequences from different species Ab initio gene-finding (intrinsic) Integrated approaches

17. Similarity-based methods Based on sequence conservation due to functional constraints Use local alignment tools (Smith-Waterman algo, BLAST, FASTA) to search protein, cDNA, and EST databases Will not identify genes that code for proteins not already in databases (can identify ~50% new genes) Limits of the regions of similarity not well defined

18. Comparative Genomics Based on the assumption that coding sequences are more conserved than non-coding Two approaches: –intra-genomic (gene families) –inter-genomic (cross-species) Alignment of homologous regions Difficult to define limits of higher similarity Difficult to find optimal evolutionary distance (pattern of conservation differ between loci)

20. Summary for Extrinsic Approaches Strengths: Rely on accumulated pre-existing biological data, thus should produce biologically relevant predictions Weaknesses: Limited to pre-existing biological data Errors in databases Difficult to find limits of similarity

21. Ab initio Gene Finding Input: A DNA string over the alphabet {A,C,G,T} Output: An annotation of the string showing for every nucleotide whether it is coding or non-coding AAAGCATGCATTTAACGAGTGCATCAGGACTCCATACGTAATGCCG Gene finder Using only sequence information Identifying only coding exons of protein-coding genes (transcription start site, 5’ and 3’ UTRs are ignored) Integrates coding statistics with signal detection

22. A eukaryotic gene This is the human p53 tumor suppressor gene on chromosome 17. Genscan is one of the most popular gene prediction algorithms. This particular gene lies on the reverse strand. 3’ untranslated region Final exon Initial exon Introns Internal exons

23. Observations Given (walk, shop, clean) –What is the probability of this sequence of observations? (is he really still at home, or did he skip the country) –What was the most likely sequence of rainy/sunny days?

24. Signals vs contents In gene finding, a small pattern within the genomic DNA is referred to as a signal, whereas a region of genomic DNA is a content. Examples of signals: splice sites, starts and ends of transcription or translation, branch points, transcription factor binding sites Examples of contents: exons, introns, UTRs, promoter regions

25. The CpG island problem Methylation in human genome –“CG” -> “TG” happens in most places except “start regions” of genes and within genes – CpG islands = 100-1,000 bases before a gene starts Question –Given a long sequence, how would we find the CpG islands in it?

26. Promoters Promoters are DNA segments upstream of transcripts that initiate transcription Promoter attracts RNA Polymerase to the transcription start site 5’ Promoter 3’

27. Splice signals (mice): GT, AG

28. Splice site detection 5’ 3’ Donor site Position %

29. Real splice sites Real splice sites show some conservation at positions beyond the first two. We can add additional arrows to model these states. weblogo.berkeley.edu

30. Ribosomal Binding Site

31. Prior knowledge The translated region must have a length that is a multiple of 3. Some codons are more common than others. Exons are usually shorter than introns. The translated region begins with a start signal and ends with a stop codon. 5’ splice sites (exon to intron) are usually GT; 3’ splice sites (intron to exon) are usually AG. The distribution of nucleotides and dinucleotides is usually different in introns and exons.

32. Gene Prediction and Motifs Upstream regions of genes often contain motifs that can be used for gene prediction -10 STOP 010-35 ATG TATACT Pribnow Box TTCCAAGGAGG Ribosomal binding site Transcription start site

33. Positional dependence In this data, every time a “G” appears in position 1, an “A” appears in position 3. Conversely, an “A” in position 1 always occurs with a “T” in position 3. ACTGACTTGCACACTTACTAGCATACTAACTTACTGACTTGCACACTTACTAGCATACTAACTT

34. Example of (Positional) Weight Matrix Computed by measuring the frequency of every element of every position of the site (weight) Score for any putative site is the sum of the matrix values (converted in probabilities) for that sequence (log-likelihood score) Disadvantages: –cut-off value required –assumes independence between adjacent bases TACGAT TATAAT GATACT TATGAT TATGTT 123456 A 060340 C 001010 G 100300 T 505016

35. Conditional probability What is the probability of observing an “A” at position 2, given that we observed a “C” at the previous position? GCG CAG CCG GCG CCG CCG GCG CCT CCG GGG CGG GCG AGG CAG CCT CAT CCT GCG

36. Conditional probability What is the probability of observing an “A” at position 2, given that we observed a “C” at the previous position? Answer: total number of CA’s divided by total number of C’s in position 1. 3/11 = 27% Probability of observing CA = 3/18 = 17%. GCG CAG CCG GCG CCG CCG GCG CCT CCG GGG CGG GCG AGG CAG CCT CAT CCT GCG

37. Conditional probability What is the probability of observing a “G” at position 3, given that we observed a “C” at the previous position? GCG CAG CCG GCG CCG CCG GCG CCT CCG GGG CGG GCG AGG CAG CCT CAT CCT GCG

38. Conditional probability What is the probability of observing a “G” at position 3, given that we observed a “C” at the previous position? Answer: 9/12 = 75%. GCG CAG CCG GCG CCG CCG GCG CCT CCG GGG CGG GCG AGG CAG CCT CAT CCT GCG

39. Promoter Structure in Prokaryotes (E.Coli) Transcription starts at offset 0. Pribnow Box (-10) Gilbert Box (-30) Ribosomal Binding Site (+10)

40. Detect potential coding regions by looking at ORFs –A genome of length n is comprised of (n/3) codons –Stop codons break genome into segments between consecutive Stop codons –The subsegments of these that start from the Start codon (ATG) are ORFs ORFs in different frames may overlap Genomic Sequence Open reading frame ATGTGA Open Reading Frames (ORFs)

41. Long open reading frames may be a gene. At random, we should expect one stop codon every (64/3) ~= 21 codons. However, genes are usually much longer than this A basic approach is to scan for ORFs whose length exceeds certain threshold. This is naïve because some genes (e.g. some neural and immune system genes) are relatively short Long vs.Short ORFs

42. Testing ORFs: Codon Usage Create a 64-element hash table and count the frequencies of codons in an ORF Amino acids typically have more than one codon, but in nature certain codons are more in use Uneven use of the codons may characterize a real gene This compensate for pitfalls of the ORF length test

43. Open Reading Frames in Bacteria Without introns, look for long open reading frame (start codon ATG, …, stop codon TAA, TAG, TGA) Short genes are missed (<300 nucleotides) Shadow genes (overlapping open reading frames on opposite DNA strands) are hard to detect Some genes start with UUG, AUA, UUA and CUG for start codon Some genes use TGA to create selenocysteine and it is not a stop codon

44. Coding Statistics Unequal usage of codons in the coding regions is a universal feature of the genomes –uneven usage of amino acids in existing proteins –uneven usage of synonymous codons (correlates with the abundance of corresponding tRNAs) We can use this feature to differentiate between coding and non-coding regions of the genome Coding statistics - a function that for a given DNA sequence computes a likelihood that the sequence is coding for a protein

45. Coding Statistics Many different ones –codon usage –hexamer usage –GC content –compositional bias between codon positions –nucleotide periodicity –…

46. Codon Usage in Human Genome

47. AA codon /1000 frac Ser TCG 4.31 0.05 Ser TCA 11.44 0.14 Ser TCT 15.70 0.19 Ser TCC 17.92 0.22 Ser AGT 12.25 0.15 Ser AGC 19.54 0.24 Pro CCG 6.33 0.11 Pro CCA 17.10 0.28 Pro CCT 18.31 0.30 Pro CCC 18.42 0.31 AA codon /1000 frac Leu CTG 39.95 0.40 Leu CTA 7.89 0.08 Leu CTT 12.97 0.13 Leu CTC 20.04 0.20 Ala GCG 6.72 0.10 Ala GCA 15.80 0.23 Ala GCT 20.12 0.29 Ala GCC 26.51 0.38 Gln CAG 34.18 0.75 Gln CAA 11.51 0.25 Codon Usage in Mouse Genome

48. Codon Usage and Likelihood Ratio An ORF is more “believable” than another if it has more “likely” codons Do sliding window calculations to find ORFs that have the “likely” codon usage Allows for higher precision in identifying true ORFs; much better than merely testing for length. However, average vertebrate exon length is 130 nucleotides, which is often too small to produce reliable peaks in the likelihood ratio Further improvement: in-frame hexamer count (frequencies of pairs of consecutive codons)

49. Splicing Signals Try to recognize location of splicing signals at exon-intron junctions. This has yielded a weakly conserved donor splice site and acceptor splice site Profiles for sites are still weak, and lends the problem to the Hidden Markov Model (HMM) approaches, which capture the statistical dependencies between sites

50. Donor and Acceptor Sites: GT and AG dinucleotides The beginning and end of exons are signaled by donor and acceptor sites that usually have GT and AC dinucleotides Detecting these sites is difficult, because GT and AC appear very often exon 1exon 2 GTAC Acceptor Site Donor Site

51. A more realistic (and complex) HMM model for Gene Prediction (Genie)

52. Assessing performance: Sensitivity & Specificity Testing of predictions is performed on sequences where the gene structure is known Sensitivity is the fraction of known genes (or bases or exons) correctly predicted –“Am I finding the things that I’m supposed to find” Specificity is the fraction of predicted genes (or bases or exons) that correspond to true genes –“What fraction of my predictions are true?” In general, increasing one decreases the other

53. Measures of Prediction Accuracy, Part 1 Nucleotide level accuracy Sensitivity= Specificity= TN FP FNTN TPFN TP FN REALITY PREDICTION number of correct exons number of actual exons number of correct exons number of predicted exons

54. Measures of Prediction Accuracy, Part 2 Exon level accuracy REALITY PREDICTION WRONG EXON CORRECT EXON MISSING EXON

55. Graphic View of Specificity and Sensitivity

56. Quantifying the tradeoff: Correlation Coefficient

57. Examples of Gene Finders FGENES – linear DF for content and signal sensors and DP for finding optimal combination of exons GeneMark – HMMs enhanced with ribosomal binding site recognition Genie – neural networks for splicing, HMMs for coding sensors, overall structure modeled by HMM Genscan – WM, WA and decision trees as signal sensors, HMMs for content sensors, overall HMM HMMgene – HMM trained using conditional maximum likelihood Morgan – decision trees for exon classification, also Markov Models MZEF – quadratic DF, predict only internal exons

58. Ab initio Gene Finding is Difficult Genes are separated by large intergenic regions Genes are not continuous, but split in a number of (small) coding exons, separated by (larger) non- coding introns –in humans coding sequence comprise only a few percents of the genome and an average of 5% of each gene Sequence signals that are essential for elucidation of a gene structure are degenerate and highly unspecific Alternative splicing Repeat elements (>50% in humans) – some contain coding regions

59. Problems with Ab initio Gene Finding No biological evidence In long genomic sequences many false positive predictions Prediction accuracy high, but not sufficient

60. Integrated Approaches for Gene Finding Programs that integrate results of similarity searches with ab initio techniques (GenomeScan, FGENESH+, Procrustes) Programs that use synteny between organisms (ROSETTA, SLAM) Integration of programs predicting different elements of a gene (EuGène) Combining predictions from several gene finding programs (combination of experts)

61. Combining Programs’ Predictions Set of methods used and they way they are integrated differs between individual programs Different programs often predict different elements of an actual gene they could complement each other yielding better prediction

62. Related Work This approach was suggested by several authors Burset and Guigó (1996) –Investigated correlation between 9 gene-finding programs –99% of exons predicted by all programs were correct –1% of exons completely missed by all programs Murakami and Tagaki (1998) –Five methods for combining the prediction by 4 gene-finding programs –Nucleotide level accuracy measures improved by 3-5% in comparison with the best single

63. AND and OR Methods exon 1 exon 2 union intersection

64. Combining Genscan and HMMgene High prediction accuracy as well as reliability of their exon probability made them the best candidates for our study Genscan predicted 77% of exons correctly, HMMgene 75%, both 87% 11162491GenscanHMMgene