ctlab / true-dist-infer

TruEst: the estimator of true evolutionary distance under the INFER model

Requirements

You need installed Python 3 and have bash alias python for it. Also, some libraries are required: numpy, scipy, networkx, pygraphviz.

About

This repository contains the implementation of the INFER model and the supplementary code.

Breakpoint Graphs

All graphs are presented in graphs package and have methods for counting required statistics and drawing itself such as:

• p_even() — number of even paths in graph;
• p_odd() — number of odd paths in graph;
• p_m(m) — number of m-paths in graph;
• chr() — estimated number of chromosomes in graph;
• c() — number of cycles paths in graph;
• c_m(m) — number of m-cycles paths in graph;
• n() — number of all blocks in graph;
• d() — the minimal DCJ distance;
• b() — is half the number of breakpoints;
• save_pygraphviz(filename) — save graph to filename file, you can specify .svg exension of file for getting an image or do not specify for getting file in pygraphviz format.

Cyclic genomes

Cyclic genome graphs are presented with CyclicGenomeGraph class in file cyclic_genome_graph.py file and can be used for simulate genome rearrangements with do_k2_break() method.

Parameter n need to be passed to the constructor. Also, initial distribution can be passed by specifying distribution from scipy.stats library and it's params for its params.

Examples:

from src.graphs.cyclic_genome_graph import CyclicGenomeGraph

# Cyclic genome graph with `1000` fragile regions in initial state
g = CyclicGenomeGraph(n=1000) 
g.do_k2_break()

from src.graphs.cyclic_genome_graph import CyclicGenomeGraph

# Cyclic genome graph with `1000` fragile regions in initial state with gamma distribution on edges
g = CyclicGenomeGraph(n=1000, distribution="gamma", params=[1/3, 0, 1])
g.do_k2_break()

Linear genomes

Linear genome graphs are presented with LinearGenomeGraph class in file linear_genome_graph.py file. It's completely the same as a CyclicGenomeGraph in usage, but also need to specify the initial number of chromosomes in the constructor.

Examples:

from src.graphs.linear_genome_graph import LinearGenomeGraph

# Cyclic genome graph with `1000` fragile regions and `10` chromosomes in initial state
g = LinearGenomeGraph(n=1000, chrs=10) 
g.do_k2_break()

from src.graphs.linear_genome_graph import LinearGenomeGraph

# Cyclic genome graph with `1000` fragile regions and `10` chromosomes in initial state with gamma distribution on edges
g = LinearGenomeGraph(n=1000, chrs=10, distribution="gamma", params=[1/3, 0, 1])
g.do_k2_break()

Real data genomes

Real genomes can be analysed by RealDataGraph class from file real_data_graph.py file. Also, this class is wrapped with bash script and can be used without writing source code. So the application of real data will be described below.

Estimators

All estimators are presented in estimators package and have methods:

• predict_k(g) — pass graph from graphs package for prediction of k parameter;
• predict_n(g) — pass graph from graphs package for prediction of n parameter;
• predict(g) — for predicting both previous.

Uniform Estimator

Uniform estimator is presented with with UniformDBEstimator class in uniform_db_estimator.py file. This estimator assumes that probabilities are uniformly distributed.

Flat Dirichlet Estimator

Flat dirichlet estimator is presented with with FlatDirichletDBEstimator class in flat_dirichlet_estimator.py file. This estimator assumes that probabilities are distributed with flat Dirichlet distribution.

Original Flat Dirichlet Estimator

Implemented flat dirichlet estimator from original paper is presented with with TannierEstimator class in tannier_dbc2_estimator.py file and also requires to pass c2 parameter to predict*() methods. This estimator assumes that probabilities are distributed with flat Dirichlet distribution.

Non-flat Dirichlet Estimator

Non-flat dirichlet estimator is presented with with DirichletDBEstimator class in dirichlet_db_estimator.py file. This estimator assumes that probabilities are distributed with Dirichlet distribution with parameter alpha. You need to pass alpha to constructor.

Corrected Non-flat Dirichlet Estimator

Corrected non-flat dirichlet estimator is presented with with CorrectedDirichletDBEstimator class in dirichlet_db_estimator.py file. This estimator assumes that probabilities are distributed with Dirichlet distribution with parameter alpha and also applies the correction for the linear case. You need to pass alpha to constructor.

Usage examples

This project has some code example in the src folder and wrapped files for using real data. You can run it by yourself with your favorite IDE with specifying the working directory as the project directory or using bash scripts.

Real Data

Estimating with real data is provided with real_data_est_grimm.py and real_data_est_infercars.py python files. You can pass parameters to this file by yourself or use bash scripts. Also you can always use -h option to get help.

Real Data in grimm Format

Real data in grimm format can be estimated with run_estimate_grimm.sh script.

You can pass -h/--help option to get help:

$ ./run_estimate_grimm.sh -h
usage: real_data_est_grimm.py [-h] --file1 FILE1 --file2 FILE2 [--alpha ALPHA]
                              [--stats]

Construct a breakpoint graph with real data provided in grimm format. Then
calculates the necessary statistics and estimate true evolutionary distance
using different approaches.

required arguments:
  --file1 FILE1, -f1 FILE1
                        Path to first file in grimm format
  --file2 FILE2, -f2 FILE2
                        Path to second file in grimm format

optional arguments:
  --alpha ALPHA, -a ALPHA
                        Set alpha to a specific value. Float, default value is
                        0.33
  --stats, -s           Show main statistics of the graph: b, d, p_even, p_odd
                        and paths and cycles count

Input examples:

./run_estimate_grimm.sh --file1 example_data/H.gen --file2 example_data/M.gen

Specifying alpha for the Dirichlet estimator:

./run_estimate_grimm.sh --file1 example_data/H.gen --file2 example_data/M.gen --alpha 0.5

Getting additional statistics:

./run_estimate_grimm.sh --file1 example_data/H.gen --file2 example_data/M.gen --stats

Real Data in infercars Format

Real data in infercars format can be estimated with run_estimate_infercars.sh script.

You can pass -h/--help option to get help:

$ ./run_estimate_infercars.sh -h
usage: real_data_est_infercars.py [-h] --file FILE --species1 SPECIES1
                                  --species2 SPECIES2 [--fit_alpha]
                                  [--alpha ALPHA] [--stats]

Construct a breakpoint graph with real data provided in infercars format. Then
calculates the necessary statistics and estimate true evolutionary distance
using different approaches.

required arguments:
  --file FILE, -f FILE  Path to file in infercars format
  --species1 SPECIES1, -s1 SPECIES1
                        Name of first species
  --species2 SPECIES2, -s2 SPECIES2
                        Name of second species

optional arguments:
  --fit_alpha, -fa      Fit alpha
  --alpha ALPHA, -a ALPHA
                        Set alpha to a specific value. Float value, default is
                        0.33
  --stats, -s           Show main statistics of the graph: b, d, p_even, p_odd
                        and paths and cycles count

Input examples:

Minimal working example:

./run_estimate_infercars.sh --file example_data/Conserved.Segments --species1 hg19 --species2 mm10

You can specify alpha and get additional statistics as in grimm format:

./run_estimate_infercars.sh --file example_data/Conserved.Segments --species1 hg19 --species2 mm10 --alpha 0.5 --stats

Or fit alpha for the Dirichlet estimator by passing --fit_alpha/-fa option.

./run_estimate_infercars.sh --file example_data/Conserved.Segments --species1 hg19 --species2 mm10 --fit_alpha

Output example:

Uniform estimator, k: 294.52810322545884
Flat Dirichlet estimator, k: 302.5027390481186
Dirichlet estimator with alpha=0.3333333333333333, k: 307.70047562385764
Corrected dirichlet estimator with alpha=0.3333333333333333, k: 307.70047562385764

Estimation After k Breaks

Estimating after k breaks in simulations example is provided in estimation_example.py python file and run_etimation_example.sh bash script. Also, you can take a look at the source code of that file and easily change used graph or number of steps.

Draw Graph Example

Draw example is provided in draw_example.py python file and run_drawer_example.sh bash script. Also, you can look at the source code of that file, it's pretty easy and you can draw any breakpoint graph. Here is example of output picture.