Inferring Functional Information from Domain co-evolution Yohan Kim, Mehmet Koyuturk, Umut Topkara, Ananth Grama and Shankar Subramaniam Gaurav Chadha Deepak Desore Layout Motivation Computational Methods and Algorithms Results Conclusion Questions Motivation (1 of 2..) Prior Work Focused on understanding Protein function at the level of entire protein sequences Assumption: Complete Sequence follows single evolutionary trajectory It is well known that a domain can exist in various contexts, which invalidates the above assumption for multi-domain protein sequences Motivation (2 of 2 ..) Our approach Improvement of Multiple Profile method Constructs Co-evolutionary Matrix to assign phylogenetic similarity scores to each protein pair Identifies Co-evolving regions using residuelevel conservation Computational Methods & Algorithms Constructing phylogenetic profiles Protein(single) phylogenetic profiles Segment(Multiple) phylogenetic profiles Residue phylogenetic profiles Computing Co-evolutionary matrices Deriving phylogenetic similarity scores Protein phylogenetic profiles Phylogenetic profile is a vector which tells about the existence of a protein in a genome. Let P = {P1,P2,…,Pn} be the set of proteins and, G = {G1,G2,…,Gm} be the set of Genomes Every row represents binary phylogenetic profile of a protein. Protein phylogenetic profiles(contd.) Single phylogenetic profile ψi for protein Pi is, ψi(j) = -1 , 1 <= j <= m log(Eij) where Eij is minimum BLAST E-value of local alignment between Pi and Gj Advantage: gives degree of sequence divergence Protein phylogenetic profiles(contd.) Mutual Information I(X,Y) defined as, I(X,Y) = H(X) + H(Y) – H(X,Y), where H(X), Shannon Entropy of X is defined as, H(X) = ∑ px * log(px), xЄX and px = P[X = x] Phylogenetic similarity between ψi(j) and ψi(j) is, μs(Pi,Pj) = I(ψi, ψi) Segment phylogenetic profiles Single profile based methods could miss significant interactions. Domain D12 of P2 follows evolutionary trajectory similar to P1 and P3 which single profile method didn’t capture. Segment phylogen. profiles(contd.) Dividing each protein Pi into fixed size segments S1i,S2i,…,Ski Phylogenetic similarity between two proteins, μM(Pi,Pj) = max I(ψsi, ψtj), s,t where ψsi is phylogenetic profile of segment Ski of protein Pi Residue phylogenetic profiles Problem with multiple phylogenetic profiles: Both domains covered together by the segment S22, overriding their individual phylogenetic profiles. Significant local alignment between two proteins corresponds to the residues covered in the alignment rather than the whole sequences. Residue phylog. profiles(contd.) A(Pi,Gj) – set of significant local alignments between Protein Pi and Genome Gj T(A) = [rb,re] – interval of residues on Pi corresponding to each alignment A Є A(Pi,Gj) For each residue r on Pi phylogenetic profile is ψri(j) = min -1 , 1 <= j <= m AЄA log(E(A)) Ar = {A Є A(Pi,Gj): r Є T(A)} is the set of local alignments that contain r r Computing co-evolutionary matrices For each protein pair Pi and Pj with lengths li and lj, co-evolutionary matrix entry Mij(r,s) is, Mij(r,s) = I (ψri, ψsj), where 1 <= r <= li and 1 <= s <= lj The Co-evolutionary Matrix contains Information about which regions of the two proteins coevolved The co-evolved domain(s) appear as a block of high mutual information scores in the matrix Deriving phylogenetic similarity scores Phylogenetic similarity scores between two proteins Pi and Pj is, μC(Pi,Pj) = max 1<= r <= li 1<= s <= lj min r <= a <= r + W s <= a <= s + W Mij(a,b) where W is the window parameter that quantifies the minimum size of the region on a protein to be considered as a conserved domain. Results Implemented and tested on 4311 E.coli proteins 152 Genomes(131 Bacteria,17 Archaea,4 Eukaryota) Value of f (down-sampling factor) = 30, W = 2 These values translate in overlapping segments of 60 residue long Excluded homologous proteins from analysis Define p-value as fraction of non-homologous protein pairs (N) Results (contd.) MIS – Mutual Information Score PP – No. of predicted protein pairs PPV = TP / (TP + FP) For all μ*, coverage = TP + FP TN and FN are the no. of protein pairs that do not meet the threshold Results (contd.) Co-evolutionary matrix has 1.5 times greater coverage at PPV = 0.7 than the single profile method At same no. of PP, Co-evolutionary matrix has better PPV and sensitivity values than single profile method Results (contd.) Mutual Information score distribution for interacting and non-interacting protein pairs At 0 MIS, SP shows a peak while CM doesn’t. In other ways, at low MIS scores, SP scores over CM Results (contd.) Shows p-values of Single Profile method v/s Co-evolutionary Matrix method Scattered circles show that the two methods can predict very differently Results (contd.) – Phosphotransferase system Domain IIA(residues 1-170) and domain IIB(residue 170-320) Darker region shows that the domains have co-evolved. So we can conclude that IIB evolved with IIC rather than IIA Top-20 predicted interacting partners of protein IIAB for both methods Results (contd.) - Chemotaxis N-terminus of CheA(residues 1-200) and C-terminus of CheA(residues 540-670) co-evolved with Cterminus region of CheB (residues 170-340) Top-20 predicted interacting partners of protein CheA using both methods Results (contd.) – Kdp System N-terminal domain of KdpD (residues 1-395) co-evolved with KdpC Top-10 predicted interacting partners of protein KdpD using both methods Conclusion Results in this paper strongly suggest that co- evolution of proteins should be captured at the domain level Because domains with conflicting evolutionary histories can co-exist in a single protein sequence Regions that are important for supporting both functional and physical interactions between proteins can be detected Questions Thank You !!