Machine learning of microbial pan-genomes

Computational platform applied to large-scale M. tuberculosis antimicrobial resistance (AMR) dataset, as described in,

ES. Kavvas, E. Catoui, N. Mih, JT. Yurkovich, Y. Seif, N. Dillon, D. Heckmann, A. Anand, L. Yang, V. Nizet, JM. Monk, BO. Palsson Machine learning and structural analysis of Mycobacterium tuberculosis pan-genome identifies genetic signatures of antibiotic resistance, Nature Communications, (2018) 9:4306

Installation

git clone https://github.com/erolkavvas/microbial_AMR_ML.git

Primary scripts

• 01_pairwise_tests.ipynb
  • Determines pairwise associations between pan-genome alleles and labeled phenotypes.
  • Generates Supplementary Data File 1
• 02_ML_ensemble_SVM.ipynb
  • Performs machine learning (ensemble support vector machine) for selecting groups of alleles that are predictive of the labeled phenotypes.
  • Generates Supplementary Data File 2, Supplementary Data File 3, and svm_ensemble_data
• 03_epistatic_analysis.ipynb
  • Uses the data generated by 02_ML_ensemble_SVM.ipynb to select an initial set of gene-gene pairs, and then performs gene-gene logistic regression modeling of these gene-gene pairs to identify statistical significant genetic interactions.
  • Generates cooccurence_table_excel, cooccurence_table_figures, and Supplementary Data File 4

Primary data structures

The following dataframes are required inputs for the computational platform.

• cluster_info.csv

	clust_to_rv	gene_name	ortho	cog	product	refseq	count	score	name_to_rv	pan
Cluster 0	Rv2048c	pks12	653045.Strvi_4160	Q	Polyketide synthase	AN47_01827	1590	7958.6	0	Core
Cluster 1	Rv3344c	PE_PGRS49	0	0	PE-PGRS family protein	X171_03503	794	0.0	0	Acces
...	...	...	...	...	...	...	...	...	...	...

• pangen_allele_df.csv

Genome ID	...	Cluster0_16	Cluster0_17	...
1010834_3	...	1		...
1010835_3	...	1		...
1010836_3	...		1	...
...	...	...	...	...

• pangen_cluster_df.csv

Genome ID	Cluster 0	Cluster 1	Cluster 2	...
1438838_3	1	1	0	...
1408941_4	1	1	0	...
1422035_3	1	0	0	...
...	...	...	...	...

• resistance_data.csv

genome_id	isoniazid	rifampicin	ethambutol	...
1295764_3	R	R	R	...
1423468_3	R	R	S	...
...	...	...	...	...

External packages of note

• patsy [https://patsy.readthedocs.io/en/latest/] - for logistic regression modeling.
• entropy_estimators.py [https://github.com/gregversteeg/NPEET] - for pairwise association analysis.