A Talk on Protein Homology Detection by HMM-HMM comparisons[1] - PowerPoint PPT Presentation

A Talk on Protein Homology Detection by HMM-HMM comparisons[1] Söding, J Qing Ye Department of Computer Science University of Illinois Urbana-Champaign March 15, 2017 1 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 1/16

Introduction • Homology detection and sequence alignment are essential for protein structure prediction, function prediction and evolution. • HHSearch ◦ Generalized sequence alignment with a profile HMM to pairwise HMM alignment ◦ Included predicted secondary structure to further improve the sensitivity ◦ The higher sensitivity also leads to increased alignment quality 2 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 2/16

Log-sum-of-odds score Generalize to pairwise HMM alignment • Note that the equation is very similar to that of sequence-HMM alignment • Here, p ( x 1 , ...x L | Null ) is simply the fixed amino acid background frequency S LSO = P ( x 1 , ..., x L | co-emission on path ) P ( x 1 , ..., x L | Null ) 3 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 3/16

Notations • q i ( k ) and p j ( k ) are the emission probability at i and j th position • f ( a ) is the background frequency • P tr is the product of all transition probability through p and q . ∑ S LSO = S aa ( q i ( k ) , p k ( k )) + log P tr k : X k Y k = MM 20 q i ( a ) p j ( a ) ∑ S aa ( q i , p j ) = log f ( a ) a =1 4 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 4/16

Pairwise alignment of HMM 5 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 5/16

Pairwise alignment of HMM State Transfer • Match and Insert states emit amino acid while Delete state does not ◦ Match and Insert cannot be aligned with Delete ◦ Delete can only aligned to Delete state or Gap state ◦ Gap denotes that the column aligned with the Gap does not have homologous partner 6 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 6/16

Pairwise alignment of HMM S MM ( i, j ) = S aa ( q i , p j )  S MM ( i − 1 , j − 1) + log [ q i − 1 ( M, M ) p j − 1 ( M, M )]     S MI ( i − 1 , j − 1) + log [ q i − 1 ( M, M ) p j − 1 ( I, M )]      + max S IM ( i − 1 , j − 1) + log [ q i − 1 ( I, M ) p j − 1 ( M, M )]   S DG ( i − 1 , j − 1) + log [ q i − 1 ( D, M ) p j − 1 ( M, M )]      S GD ( i − 1 , j − 1) + log [ q i − 1 ( M, M ) p j − 1 ( D, M )]   7 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 7/16

Pairwise alignment of HMM  S MM ( i − 1 , j ) + log [ q i − 1 ( M, M ) p j ( M, I )]  S MI ( i, j ) = max S MI ( i − 1 , j ) + log [ q i − 1 ( M, M ) p j ( I, I )]   S MM ( i − 1 , j ) + log [ q i − 1 ( M, D )]  S DG ( i, j ) = max S DG ( i − 1 , j ) + log [ q i − 1 ( D, D )]  8 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 8/16

A few tricks are added to improve the performance • S aa offset = ⇒ produce shorter alignment • Sequence weighting and pseudo-count = ⇒ alignment of sequences of several subdomains transform into profile as if the alignment was cut into subdomains first and then calcualte the profile • Scoring correlation = ⇒ to expect high column scores to appear in clusters 9 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 9/16

Scoring secondary structure Compare against actual secondary structure • σ is a DSSP state • ρ is a PSIPred state • c is the confidence value P ( σ ; ρ, c ) M SS ( σ ; ρ, c ) = log P ( σ ) P ( ρ, c ) 10 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 10/16

Compare against predicted secondary structure • Runs over all DSSP state P ( ρ q i , c q i | σ ) P ( ρ q j , c q j | σ ) M SS ( ρ q i , c q i ; ρ q j , c q ∑ j ) = log P ( σ ) P ( ρ q i , c q i ) P ( ρ q j , c q j ) σ 11 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 11/16

Results 12 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 12/16

Results 13 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 13/16

Results 14 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 14/16

Results 15 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 15/16

Reference I [1] J. Söding. Protein homology detection by hmm–hmm comparison. Bioinformatics, 21(7):951, 2005. 16 / 16 Qing Ye qingye3@illinois.edu Protein homology detection by HMM-HMM comparison 16/16