model_type # 0, single-end, no quality score; 1, single-end, quality score; 2, paired-end, no quality score; 3, paired-end, quality score forward_prob # The probability of generating a read from forward strand # Fragment length distribution or read length distribution (if reads are single-end reads and no fragment length distribution parameters are provided) lower_bound upper_bound span # Fragment/read length should be in the range of (lower_bound, upper_bound] and span = upper_bound - lower_bound prob_1 prob_2 ... prob_span # This line contains span probabilities separated by a single space. prob_i is the probability of generating a length of (lower_bound + i) [ # Read length distribution, this section is optional for single-end reads model, the "[" and "]" suggest this section is optional. has_optional_length_dist # This line only appear if model_type < 2. 0 stands no optional length distribution for that the previous section already defines the read length distribution, 1 stands for having this optional read length distribution. If has_optional_length_dist = 0, the next two lines will not appear lower_bound upper_bound span # Fragment/read length should be in the range of (lower_bound, upper_bound] and span = upper_bound - lower_bound prob_1 prob_2 ... prob_span # This line contains span probabilities separated by a single space. prob_i is the probability of generating a length of (lower_bound + i) ] # Read start position distribution estimate_RSPD # 0 means that RSEM assumes a uniform read generating distribution across a transcript, 1 means that RSEM estimated a RSPD from data number_of_bins # number of bins used for estimating RSPD, please refer to Li, B., Ruotti, V., Stewart, R. M., Thomson, J. A., Dewey, C. N. (2010). RNA-Seq gene expression estimation with read mapping uncertainty. Bioinformatics, 26(4), 493-500. for more details bin_prob_1 bin_prob_2 ... bin_prob_number_of_bins # number_of_bins single space separated numbers representing the probabilities of starting a read from each bin [ # Quality score distribution, only presented if model_type == 1 or 3. Quality scores are generated according to a first order Markov model # Quality scores are numbered from 0 internally, thus if prhed+33 quality score is used in the data, then i corresponds to phred score (i+33) size # total number of quality scores modeled p_init_0 p_init_1 ... p_init_(size-1) # size real-valued, single space separated numbers representing the initial probabilities of each state (quality score) p_trans_0,0 ... p_trans_0,(size-1) ... p_trans_(size-1),0 ... p_trans_(size-1),(size-1) # size times size matrix representing the transition matrix for the Morkov chain ] [ # Sequencing error model for quality scores, only presented if model_type == 1 or 3 size ncodes # size is the total number of quality scores and ncodes is the number of different bases (ncodes = 5, 0: A; 1: C; 2: G; 3: T; 4: N) # In the following, RSEM gives size blocks. The i th block represents the sequencing error model for quality score i. Blocks are separated by single blank lines ... p_i,0,0 p_i,0,1 p_i,0,2 p_i,0,3 p_i,0,4 ... p_i,4,0 p_i,4,1 p_i,4,2 p_i,4,3 p_i,4,4 # The i th block contains a ncodes times ncodes matrix. Each cell (j,k) gives the probability of generating a base k given the quality score is i and the reference base is j ... ] [ # Model for generating a noise read based on quality scores. Only presented if model_type == 1 or 3 size ncodes # the same as sequencing error model for quality score p_0,0 p_0,1 p_0,2 p_0,3 p_0,4 ... p_(size-1),0 p_(size-1),1 p_(size-1),2 p_(size-1),3 p_(size-1),4 # A size times ncode matrix giving the probability of generating a base in a noise read given a quality score ] [ # Sequencing error model without quality score, only presented if model_type == 0 or 2. It records the probability of generating a base k at position i given the corresponding reference base is j. Positions are numbered from 0 profile_length ncodes # profile_length should be equal to the maximum length of any read in the data set # There will be profile_length blocks separated by single blank lines ... p_i,0,0 p_i,0,1 p_i,0,2 p_i,0,3 p_i,0,4 ... p_i,4,0 p_i,4,1 p_i,4,2 p_i,4,3 p_i,4,4 # The i th block contains a ncodes times ncodes matrix. Each cell (j,k) gives the probability of generating a base k at position i given the reference base is j ... ] [ # Model for generating a noise read without quality scores. Only presented if model_type == 0 or 2 ncodes # ncodes = 5 p_0 p_1 p_2 p_3 p_4 # probabilities of generating a particular base for the noise read, all positions use the same base generating probabilities ] [ # Some other numbers, which has nothing to do with simulation ]