ó )†Wc@sŠddlZddlmZddlmZd„Zd„Zd„Zd„Z d„Z d „Z d „Z d „Z d „Zd „ZdS(iÿÿÿÿN(tutils(t fclusterdatacCs"t|ƒt|ƒt||ƒS(N(t log_factorial(tntx((sDP_GP/cluster_tools.pytlog_binomial_coefficientscCstj|dƒS(Ni(tnptlog(R((sDP_GP/cluster_tools.pyR sc Csˆ|jd}tjt|dƒƒ}d}xTt|ƒD]F}x=t|ƒD]/}||||krK||||7}qKqKWq8Wd}xft|ƒD]X}xOt|ƒD]A}||||kr¨||d|d…|fjƒ7}q¨q¨Wq•W||:}d}x^t|ƒD]P}xGt|ƒD]9}||||7}||||kr!|d7}q!q!WqW|d:}||} ||} | | S(sÛ Compute MPEAR (Fritsch and Ickstadt 2009, DOI:10.1214/09-BA414). This function and accessory routines were taken with little modification from Pyclone (Roth et al. 2014 DOI:10.1038/nmeth.2883). :param cluster_labels: cluster labels :type cluster_labels: numpy array of ints :param sim_mat: sim_mat[i,j] = (# samples gene i in cluster with gene j)/(# total samples) :type sim_mat: numpy array of (0-1) floats :rtype: float iiNi(tshapeRtexpRtrangetsum( tcluster_labelstsim_mattNtct num_term_1tjtit num_term_2t den_term_1tnumtden((sDP_GP/cluster_tools.pyt compute_mpears. /    cCs}i}d}g}xdt|ƒD]V}||krU|j|ƒ|||<|d7}q||kr|j||ƒqqW|S(sV Given some cluster labels, relabel (equivalently) so that labels start at 1. i(tlisttappend(R t clust_dictt new_labelt new_labelstlabel((sDP_GP/cluster_tools.pytrelabel_clusteringEs     cCs)||}tjtj||ƒƒ}|S(s@ Compute the squared distance between two matrices. (RR tdot(tStS_newtdifftsq_dist((sDP_GP/cluster_tools.pytcompute_sq_distXs cCshd}xOtt|ƒƒD];}t||f|ƒ}||kr|}||}qqWt|ƒ}|S(sš Find the optimal clustering according to the MPEAR criterion. :param clusterings: clusterings[i,j] is the cluster to which gene j belongs at sample i :type clusterings: numpy array of ints :param sim_mat: sim_mat[i,j] = (# samples gene i in cluster with gene j)/(# total samples) :type sim_mat: numpy array of (0-1) floats :rtype: numpy array of best cluster labels i(R tlenRR(t clusteringsR tmax_pearRtpeartbest_cluster_labelsR((sDP_GP/cluster_tools.pytbest_clustering_by_mpearcs   cCsetj }xHttt|ƒƒ|ƒD]+\}}||kr&|}||}q&q&Wt|ƒ}|S(sÁ Find the optimal clustering according to log posterior likelihood (i.e. maximum a posteriori clustering). :param clusterings: clusterings[i,j] is the cluster to which gene j belongs at sample i :type clusterings: numpy array of ints :param log_post_list: list of log posterior likelihood over the course of Gibbs sampling :type log_post_list: list of floats :rtype: numpy array of best cluster labels (RtinftzipR R%R(R&t log_post_listtmax_postRtpostR)R((sDP_GP/cluster_tools.pyt!best_clustering_by_log_likelihoods (  c Cs tj}xðtt|ƒƒD]Ü}||dd…f}tjt|ƒt|ƒfƒ}xntt|ƒƒD]Z}xQtt|ƒƒD]=}||||kr²d|||fgene (e.g.) 1gene1 . . . s_optimal_clustering.txttws cluster gene s%s %s N(topentwritetsortedRt sorted_nicelytclose(toptimal_cluster_labelstoutput_path_prefixthandletclustertgenestgene((sDP_GP/cluster_tools.pyt#save_cluster_membership_informationÑs    (tnumpyRtDP_GPRtscipy.cluster.hierarchyRRRRRR$R*R0R6R:RG(((sDP_GP/cluster_tools.pyts    4    $