feat: add hart2015 essentiality suite

[haowang-bioinfo]

Main improvements in this PR:

This PR provides infrastructure of essentiality test suite, as proposed in #390, using Hart2015 datasets with following components:

• Hart2015_RNAseq.txt: RNA-seq sequences of Hart2015 cell lines (GEO: GSE75189)
• Hart2015_TableS2.xlsx: Calcuated Bayes factors for CRISPR targeted genes
• getTaskEssentialGenes: Identify genes essential for different tasks in different tINIT models
• estimateEssentialGenes: Generate tINIT models and estimate essential genes
• evaluateHart2015Essentiality: Evaluate and compare Hart2015 experimental fitness genes with predicted results

I hereby confirm that I have:

• Tested my code on my own computer for running the model
• Selected develop as a target branch
• Any removed reactions and metabolites have been moved to the corresponding deprecated identifier lists

[mihai-sysbio]

This is awesome! A quick thought: can the xlsx file be avoided?

[haowang-bioinfo]

A quick thought: can the xlsx file be avoided?

no, not at the moment - would like to retain the original Hart2015 data file intact

[mihai-sysbio]

retain the original Hart2015 data file intact

I understand - it makes a lot of sense. And since this file will not be modified, it's not going to continuously enlarge the repository.

[haowang-bioinfo]

I understand - it makes a lot of sense. And since this file will not be modified, it's not going to continuously enlarge the repository.

thanks - yes there's no need to change the content, any modification, such as converting into a plain-text version (which does make sense), would confuse people though

[mihai-sysbio]

infrastructure of essentiality test suite

Is the aim to run the essentiality suite manually in conjunction with every release?

[haowang-bioinfo]

Is the aim to run the essentiality suite manually in conjunction with every release?

yes, this test can be applied to evaluate each release among others

[haowang-bioinfo]

Updated essentiality evaluation using combined (all) Hart2015 datasets:

version	TP	TN	FP	FN	accuracy	sensitivity	specificity	F1	MCC
v1.12	40	2333	175	77	0.904000000000000	0.341880341880342	0.930223285486443	0.240963855421687	0.204768560393159
v1.13	40	2334	174	77	0.904380952380952	0.341880341880342	0.930622009569378	0.241691842900302	0.205504558241293
v1.14	40	2334	175	77	0.904036557501904	0.341880341880342	0.930251096054205	0.240963855421687	0.204787829413435
v1.15	40	2342	168	77	0.906737723639132	0.341880341880342	0.933067729083665	0.246153846153846	0.210053956889428
develop	41	2317	194	76	0.897260273972603	0.350427350427350	0.922739944245321	0.232954545454545	0.197442075842213

The results show that the accumulated curations since v1.15 have positive effect to TP, FN, but adversely affect TN and FP (by which PRs?)

[haowang-bioinfo]

now it might be better to spin out a new release, to present our recent work for community inspection.

it would also be good at some point to look into which PRs led to increased FP, probably by decreasing TN

[mihai-sysbio]

to present our recent work for community inspection.

To me, this is already done in the develop branch.

it would also be good at some point to look into which PRs led to increased FP, probably by decreasing TN

Why "at some point" and not now when the changes are more fresh?

[haowang-bioinfo]

to present our recent work for community inspection.

To me, this is already done in the develop branch.

the inspection is best achieved by using the model, and community access is mainly through the releases, instead of the develop branch that involves so far a small group of people. To me, I have limited knowledge to affirm every single change. The best practice probably is to ensure release early, release often in a radically transparent way

it would also be good at some point to look into which PRs led to increased FP, probably by decreasing TN

Why "at some point" and not now when the changes are more fresh?

sure go ahead please

[haowang-bioinfo]

Code example for running essentiality analysis with Hart2015 datasets:

% load model and essential tasks
load('Human-GEM.mat');  % or: ihuman = importYaml('Human-GEM.yml');
taskStruct = parseTaskList('metabolicTasks_Essential.txt');

% generate tINIT models and estimate essential genes
eGenes = estimateEssentialGenes(ihuman, 'Hart2015_RNAseq.txt', taskStruct);

% compare model predictions with experimental data
results = evaluateHart2015Essentiality(eGenes);

it may take 2-4 hours or more, depending on computer's configuration, to complete this analysis on a laptop or desktop.

[JonathanRob]

This is really nice, good work @haowang-bioinfo. I agree it's a bit unfortunate with the large-ish .xlsx file, but as @mihai-sysbio says, it won't need to be changed.

[mihai-sysbio]

Code example for running essentiality analysis with Hart2015 datasets:
it may take 2-4 hours or more, depending on computer's configuration, to complete this analysis on a laptop or desktop.

Shouldn't we add this then to an action that would run for every pull request from develop to main?

[haowang-bioinfo]

Shouldn't we add this then to an action that would run for every pull request from develop to main?

not sure if this is suitable for GH action, because the computation load is very heavy. Probably we would know better after solving #635

[feiranl]

The previous steps works fine, the last one encounters an error

>> results = evaluateHart2015Essentiality(eGenes);
Unrecognized function or variable 'adjust_pvalues'.

Error in evaluateHart2015Essentiality (line 129)
PenrAdj = adjust_pvalues(Penr,'Benjamini');

[haowang-bioinfo]

the last one encounters an error

ah, yes - now adjust_pvalues was added from this repo for convenient use

[feiranl]

Successfully run and got this result

'cellLine'	'TP'	'TN'	'FP'	'FN'	'accuracy'	'sensitivity'	'specificity'	'F1'	'MCC'	'Penr'	'logPenr'	'PenrAdj'	'logPenrAdj'
'DLD1'	95	2109	143	223	0.85758755	0.29874214	0.93650089	0.34172662	0.26721557	1.04E-30	29.9838919	3.11E-30	29.5067706
'GBM'	85	2081	153	250	0.84312962	0.25373134	0.93151298	0.29668412	0.21515393	1.78E-21	20.7503043	2.67E-21	20.5742131
'HCT116'	113	2137	130	246	0.85681645	0.31476323	0.94265549	0.37541528	0.3051743	3.02E-40	39.519913	1.81E-39	38.7417618
'HELA'	86	2182	157	202	0.86334222	0.29861111	0.9328773	0.32391714	0.24962285	2.74E-27	26.5616613	5.49E-27	26.2606313
'RPE1'	63	2115	175	216	0.8478007	0.22580645	0.92358079	0.24371373	0.16031484	6.69E-13	12.1747792	6.69E-13	12.1747792
'all'	41	2308	202	76	0.89417587	0.35042735	0.91952191	0.22777778	0.19220101	2.19E-15	14.6589606	2.63E-15	14.5797793

[feiranl]

LGTM!