Computational methods for prediction of alternative promoters and TFBS

CREATED: 200811100603 Speaker: Junwen Wang ** Structure of a gene

• Eukaryotic gene: > 1000bp
• Core promoter region: ~100bp
• TFBS: 6-20bp ** Prediction models
• Content sensor ** Markov model ** captures context information ignores position specifity ** for long sequences
• Signal sensor ** Position weight matrix ** captures position specific information but ignores context ** for short sequences ** Generalized Markov Model for long sequences
• instead of using a single nucleotide as a unit, allow di/tri nucleotide
• allow gaps
• evaluate on promoter/exon/nucleosome sequences
• in gapped model, distance between best performance ~11bp ** Position Specific Propensity Model for medium length sequences
• propensity of a k-mer using variable length k-mer
• 5’ promoter more likely to be in CpG island
• whole genome alternative promoter predictions ** Modeling of short sequences (TFBS)
• PWM assumes position independence
• PSPM or variable length MM of binding sites is superior to the PWM model