Minor feature update due to located memory bug when using long progenitor sequences in focussed mode.
1: memorylimit added to jobstart.csv.
Setting this option to YES will cause Proseeker to use a smaller number of variants than all possible variants when creating a focussed combination set. Use this option when running in a desktop environment to avoid memory errors.
2: memorythresh added to jobstart.csv.
The integer in this option specifies how many random variants Proseeker focussed mode is produced. This must be set to X>0 when memorylimit is set to YES.
3: bresnum added to jobstart.csv.
Specifies the number of BRES files to be read in the library. This number is taken as sequential (i.e. setting it to 15 will result in bres 1-15 being read even if there are greater than 15 bres files in the directory.
jobstart.csv is now copied into the results directory as used_jobstart.csv
Focus mode is now added to allow targeted, selectable mutagenesis for assistance in making libraries. To activate this mode change the fmode option in jobstart.csv to YES. Once this is done Proseeker will run a single continuous generation with using the g1 protein sequence as a template. It will make every possible combination of amino acids of length n where n = the number of mutable residues in the sequence and only amino acids having non-zero probabilities of selection will be used after which the standard assessment loop will assess each protein sequence against the library in random order.
The generation will end when sufficient evidence of normality is observed (p > 0.05 using a Shapiro-Wilks test) with assessment of normality starting once either fthresh variants have been tested or the complete pool of variants have been tested. In the case that the number of variants in the pool is lower than fthresh then no normality test will be conducted.
The output files from this mode will show the variants tested, representation of Amino Acids at mutable positions within the upper and lower quartiles and give a preliminary assessment on which residues are likely to have a negative effect when oberserved in a sequence at each mutable position.
New library requirements: scipy and itertools
Proseeker is supplied as a standalone Python script (Proseeker.py) for either interactive use or pipeline integration and was written in Python 3.9.
Packages required:
numpy
pandas
sklearn
If running in a pipeline the commenting on the below lines (38 & 40 in Proseeker.py as written) shouled be uncommented and the path to the working directory supplied as a commandline argument:
user_input = input("Enter the path to your Proseeker working directory (i.e C:\Proseeker): ")
# user_input = int(sys.argv[1])
The Proseeker working directory should contain the files ranking.csv, jobstart.csv and whatever you have named your Bres file directory (i.e. PPRONLY).
The ranking file contains values drawn from AAindex as stated in the paper.
The jobstart.csv file (example included) is laid out as below in two lines with headers as shown:
g,k,MEG,block,g1,pchoice,A,R,N,D,C,Q,E,G,H,I,L,K,M,F,P,S,T,W,Y,V,truncres,cores,sites,bres
20,100,2,MMMMFFFFFFFFMMMMMMMMMMMMMFFFFFFFFFFFFFMM,SELVNEIVKQLAEVAKEATDKELVIYIVKILAELAKQSTD,1,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,0.05,10,4,2,BRES
g = number of generations and should be any positive integer greater than 1 (20 is the default).
k = the number of variants to be amde per generation. This should be a positive integer greater than 10 (100 is the default and is recommended).
MEG = the number of mutational events to occur per generation (this differs from EMEG as stated in the paper).
block = the masking scheme to be used. M=mutable and F = fixed. This is not option so if no masking is desired simply include M at all positions. This string must be the same length as the progenitor protein.
g1 = the progenitor protein. This should contain only a string consisting of letters (case is irrelevant) corresponding to one of the 20 natural amino acids. STOP codon is not included and Proseeker has no ability to introduce STOP codons.
pchoice = the choice to use a specified probability set or not. 1 = yes and 0 = no.
A,R,N,D,C,Q,E,G,H,I,L,K,M,F,P,S,T,W,Y,V = the individual probability of selection of each AA. These should add up to 1 and as such 0.05 (20*0.05) represents an even probability of selection. If phoice = 1 then the values listed here will be used.
truncres = Proportion of results to provide to the user. The default is 10 which will results in the best k/10 ressults per generation being provided to the user.
cores = specification of number cores. This is present to support the parallelized version. This is set to 4 default however for non-cluster use it may be ignored.
sites = The number of sites per protein within the assessment library that are to be used as comparators.
BRES = the name of the directory in which your BRES files are stored (i.e. if the BRES files are in C:\path\to\Proseeker\BRES then you would specify just BRES).
Once run Proseeker will create a time-stamped directory for the run within the working directory (the jobstart.csv file is only needed initially and as such after run commencement may be moved or altered). This directory will be empty at first however it will save each generations best results in the created file for later harvesting and examination. As such multiple runs can be performed simultaneously on a desktop enironment allowing triplicate runs for verification and assessment of consistency.
Making BRES files is acomplished using the BresMaker.py script. BRES files used here are simplified BRES files.
BresMaker is supplied as a standalone Python script (BresMaker.py) for either interactive use or pipeline integration and was written in Python 3.9.
Packages required:
numpy
pandas
sklearn
This script may be run in a pipeline or in an interactive environment by changing commenting on the below lines to produce ether a commandline or interactive version (24:34 in BresMaker.py as written):
#user_input = str(sys.argv[1])
#ranking = str(sys.argv[2])
#working = str(sys.argv[3])
#iterations = int(sys.argv[4])
#trys = int(sys.argv[5])
user_input = "D:/Proseeker/exampledeets.csv"
ranking = "D:/Proseeker/ranking.csv"
working = "D:/Proseeker"
iterations = 1000000
trys = 1000
The details file (exampledeets.csv) loaded as 'user_input' consists of a .csv file with two columns as below:
LTPEQVVAIASNIGGKQALETVQALLPVLCQAHG,12:13:14
Column 1 consists of one string per row consisting of letters (case is irrelevant) corresponding to one of the 20 natural amino acids. STOP codon is not included and Proseeker has no ability to introduce STOP codons.
Column 2 consists of positive integers (you could use negative integers as Python will take -1 for example to refer to the last character however it is easy to get negative indexing wrong). If only 1 site is to be taken then 1 integer is provided. If more than 1 site is required then 1 integer per site is required separated by : with no spaces.
The ranking file (ranking.csv) loaded as 'ranking' contains values drawn from AAindex as stated in the paper.
Working is the directory in which the BresFiles are to be created. It is recommended that a specific directory be created for this.
Iterations and Trys refer to the maximum number of k-means clustering interations to be performed and the number of time the k-means algorithm will be run with different centroid seeds respectively. This process works better the tighter the clusters are. As such It is suggest that a minimum of trys = 100 should be used while the default is 1000. This will result in some extra time being spent preparing the library but produces more consistent, higher quality evolution runs so in the long run extra preparation results in higher success rates with a small initial time investment.