Gene Recognition

Gene Recognition

Gene structure intron1 intron2 exon2 exon3 exon1 transcription splicing translation Codon: A triplet of nucleotides that is converted to one amino acid exon = protein-coding intron = non-coding

Needles in a Haystack

Gene Finding EXON EXON EXON EXON EXON • Classes of Gene predictors • Ab initio • Only look at the genomic DNA of target genome • De novo • Target genome + aligned informant genome(s) • EST/cDNA-based & combined approaches • Use aligned ESTs or cDNAs + any other kind of evidence Human tttcttagACTTTAAAGCTGTCAAGCCGTGTTCTAGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtaccta Macaque tttcttagACTTTAAAGCTGTCAAGCCGTGTTCTAGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtaccta Mouse ttgcttagACTTTAAAGTTGTCAAGCCGCGTTCTTGATAAAATAAGTATTGGACAACTTGTTAGTCTTCTTTCCAACAACCTGAACAAATTTGATGAAgtatgta-cca Rat ttgcttagACTTTAAAGTTGTCAAGCCGTGTTCTTGATAAAATAAGTATTGGACAACTTATTAGTCTTCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtaccca Rabbit t--attagACTTTAAAGCTGTCAAGCCGTGTTCTAGATAAAATAAGTATTGGGCAACTTATTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtaccta Dog t-cattagACTTTAAAGCTGTCAAGCCGTGTTCTGGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTCGATGAAgtatgtaccta Cow t-cattagACTTTGAAGCTATCAAGCCGTGTTCTGGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgta-cta Armadillo gca--tagACCTTAAAACTGTCAAGCCGTGTTTTAGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtgccta Elephant gct-ttagACTTTAAAACTGTCCAGCCGTGTTCTTGATAAAATAAGTATTGGACAACTTGTCAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtatcta Tenrectc-cttagACTTTAAAACTTTCGAGCCGGGTTCTAGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtatcta Opossum ---tttagACCTTAAAACTGTCAAGCCGTGTTCTAGATAAAATAAGCACTGGACAGCTTATCAGTCTCCTTTCCAACAATCTGAACAAGTTTGATGAAgtatgtagctg Chicken ----ttagACCTTAAAACTGTCAAGCAAAGTTCTAGATAAAATAAGTACTGGACAATTGGTCAGCCTTCTTTCCAACAATCTGAACAAATTCGATGAGgtatgtt--tg

Signals for Gene Finding • Regular gene structure • Exon/intron lengths • Codon composition • Motifs at the boundaries of exons, introns, etc. Start codon, stop codon, splice sites • Patterns of conservation • Sequenced mRNAs • (PCR for verification)

Next Exon: Frame 0 Next Exon: Frame 1

Exon and Intron Lengths

Nucleotide Composition • Base composition in exons is characteristic due to the genetic code Amino Acid SLCDNA Codons Isoleucine I ATT, ATC, ATA Leucine L CTT, CTC, CTA, CTG, TTA, TTG Valine V GTT, GTC, GTA, GTG Phenylalanine F TTT, TTC Methionine M ATG Cysteine C TGT, TGC Alanine A GCT, GCC, GCA, GCG Glycine G GGT, GGC, GGA, GGG Proline P CCT, CCC, CCA, CCG Threonine T ACT, ACC, ACA, ACG Serine S TCT, TCC, TCA, TCG, AGT, AGC Tyrosine Y TAT, TAC Tryptophan W TGG Glutamine Q CAA, CAG Asparagine N AAT, AAC Histidine H CAT, CAC Glutamic acid E GAA, GAG Aspartic acid D GAT, GAC Lysine K AAA, AAG Arginine R CGT, CGC, CGA, CGG, AGA, AGG

atg caggtg ggtgag cagatg ggtgag cagttg ggtgag caggcc ggtgag tga

Splice Sites (http://www-lmmb.ncifcrf.gov/~toms/sequencelogo.html)

intron exon exon intron intergene exon intergene HMMs for Gene Recognition First Exon State Intron State Intergene State GTCAGATGAGCAAAGTAGACACTCCAGTAACGCGGTGAGTACATTAA

Exon1 Exon2 Exon3 Duration HMMs for Gene Recognition PSTOP(xi-4…xi+3) Duration d j+2 T A A T A T G T C C A C G G G T A T T G A G C A T T G T A C A C G G G G T A T T G A G C A T G T A A T G A A iPINTRON(xi | xi-1…xi-w) P5’SS(xi-3…xi+4) PEXON_DUR(d)iPEXON((i – j + 2)%3)) (xi | xi-1…xi-w)

Genscan • Burge, 1997 • First competitive HMM-based gene finder, huge accuracy jump • Only gene finder at the time, to predict partial genes and genes in both strands Features • Duration HMM • Four different parameter sets • Very low, low, med, high GC-content

Using Comparative Information

Using Comparative Information • Hox cluster is an example where everything is conserved

Genes Intergenic Separation Mutations Gaps Frameshifts 30% 1.3% 0.14% 58% 14% 10.2%    2-fold 10-fold 75-fold Patterns of Conservation

Comparison-based Gene Finders • Rosetta, 2000 • CEM, 2000 • First methods to apply comparative genomics (human-mouse) to improve gene prediction • Twinscan, 2001 • First HMM for comparative gene prediction in two genomes • SLAM, 2002 • Generalized pair-HMM for simultaneous alignment and gene prediction in two genomes • NSCAN, 2006 • Best method to-date based on a phylo-HMM for multiple genome gene prediction

Twinscan • Align the two sequences (eg. from human and mouse) • Mark each human base as gap ( - ), mismatch ( : ), match ( | ) New “alphabet”: 4 x 3 = 12 letters • = { A-, A:, A|, C-, C:, C|, G-, G:, G|, U-, U:, U| } • Run Viterbi using emissions ek(b) where b  { A-, A:, A|, …, T| } Emission distributions ek(b) estimated from real genes from human/mouse eI(x|) < eE(x|): matches favored in exons eI(x-) > eE(x-): gaps (and mismatches) favored in introns Example Human: ACGGCGACGUGCACGU Mouse: ACUGUGACGUGCACUU Alignment: ||:|:|||||||||:|

SLAM – Generalized Pair HMM Exon GPHMM 1.Choose exon lengths (d,e). 2.Generate alignment of length d+e. e d

NSCAN—Multiple Species Gene Prediction • GENSCAN • TWINSCAN • N-SCAN Target GGTGAGGTGACCAAGAACGTGTTGACAGTA Target GGTGAGGTGACCAAGAACGTGTTGACAGTA Conservation |||:||:||:|||||:||||||||...... sequence Target GGTGAGGTGACCAAGAACGTGTTGACAGTA Informant1 GGTCAGC___CCAAGAACGTGTAG...... Informant2 GATCAGC___CCAAGAACGTGTAG...... Informant3 GGTGAGCTGACCAAGATCGTGTTGACACAA ... Target sequence: Informant sequences (vector): Joint prediction (use phylo-HMM):

NSCAN—Multiple Species Gene Prediction H X C Y Y H Z X Z M R C M R

Performance Comparison NSCAN human/mouse > Human/multiple informants GENSCAN Generalized HMM Models human sequence TWINSCAN Generalized HMM Models human/mouse alignments N-SCAN Phylo-HMM Models multiple sequence evolution

CONTRAST • 2-level architecture • No Phylo-HMM that models alignments Human tttcttagACTTTAAAGCTGTCAAGCCGTGTTCTAGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtaccta Macaque tttcttagACTTTAAAGCTGTCAAGCCGTGTTCTAGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtaccta Mouse ttgcttagACTTTAAAGTTGTCAAGCCGCGTTCTTGATAAAATAAGTATTGGACAACTTGTTAGTCTTCTTTCCAACAACCTGAACAAATTTGATGAAgtatgta-cca Rat ttgcttagACTTTAAAGTTGTCAAGCCGTGTTCTTGATAAAATAAGTATTGGACAACTTATTAGTCTTCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtaccca Rabbit t--attagACTTTAAAGCTGTCAAGCCGTGTTCTAGATAAAATAAGTATTGGGCAACTTATTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtaccta Dog t-cattagACTTTAAAGCTGTCAAGCCGTGTTCTGGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTCGATGAAgtatgtaccta Cow t-cattagACTTTGAAGCTATCAAGCCGTGTTCTGGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgta-cta Armadillo gca--tagACCTTAAAACTGTCAAGCCGTGTTTTAGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtgccta Elephant gct-ttagACTTTAAAACTGTCCAGCCGTGTTCTTGATAAAATAAGTATTGGACAACTTGTCAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtatcta Tenrectc-cttagACTTTAAAACTTTCGAGCCGGGTTCTAGATAAAATAAGTATTGGACAACTTGTTAGTCTCCTTTCCAACAACCTGAACAAATTTGATGAAgtatgtatcta Opossum ---tttagACCTTAAAACTGTCAAGCCGTGTTCTAGATAAAATAAGCACTGGACAGCTTATCAGTCTCCTTTCCAACAATCTGAACAAGTTTGATGAAgtatgtagctg Chicken ----ttagACCTTAAAACTGTCAAGCAAAGTTCTAGATAAAATAAGTACTGGACAATTGGTCAGCCTTCTTTCCAACAATCTGAACAAATTCGATGAGgtatgtt--tg X SVM SVM CRF a b a b Y

CONTRAST

CONTRAST - Features • log P(y | x) ~ wTF(x, y) • F(x, y) = if(yi-1, yi, i, x) • f(yi-1, yi, i, x): • 1{yi-1 = INTRON, yi = EXON_FRAME_1} • 1{yi-1 = EXON_FRAME_1, xhuman,i-2,…, xhuman,i+3 = ACCGGT) • 1{yi-1 = EXON_FRAME_1, xhuman,i-1,…, xdog,i+1 = ACC, AGC) • (1-c)1{a<SVM_DONOR(i)<b} • (optional) 1{EXON_FRAME_1, EST_EVIDENCE}

CONTRAST – SVM accuracies SN SP • Accuracy increases as we add informants • Diminishing returns after ~5 informants

CONTRAST - Decoding Viterbi Decoding: maximize P(y | x) Maximum Expected Boundary Accuracy Decoding: maximize i,B 1{yi-1, yi is exon boundary B} Accuracy(yi-1, yi, B | x) Accuracy(yi-1, yi, B | x) = P(yi-1, yi is B | x) – (1 – P(yi-1, yi is B | x))

CONTRAST - Training Maximum Conditional Likelihood Training: maximize L(w) = Pw(y | x) Maximum Expected Boundary Accuracy Training: ExpectedBoundaryAccuracy(w) = i Accuracyi Accuracyi = B1{(yi-1, yi is exon boundary B} Pw(yi-1, yi is B | x) - B’ ≠ B P(yi-1, yi is exon boundary B’ | x)

Performance Comparison Human Macaque Mouse Rat Rabbit Dog Cow Armadillo Elephant Tenrec Opossum Chicken De Novo EST-assisted

Performance Comparison

Gene Recognition