Four strategies take a list: 02 looks for the cluster that holds it, 20 learns what makes it special from positives alone, 24 describes what it has in common, and 25 grows it along the networks. Here the list is the 135 genes of PM - integral; hold out the compartment column so no strategy can read the list back from the label that defined it.
from starplast import strategies as Sctx = S.shipped('Tg')cat, members = S.example_set(ctx, 'compartment')genes = list(ctx.gene_ids[members])print(cat, len(genes), genes[:5])
PM - integral 135 ['TGME49_254370', 'TGME49_216620', 'TGME49_247195', 'TGME49_211640', 'TGME49_316730']
hunt = S.run('geneset_hunt', genes=genes, exclude='compartment', n_neighbors='15, 50', min_dist='0.0', min_cluster_size='20, 50', selection='eom, leaf', sample=2000)print(hunt.summary)
135 genes found (0 not found). Best of 8 configurations: all permitted, n_neighbors 15, min_dist 0.0, min cluster 20 (leaf) -- cluster 5 holds 22 of your 135 placed genes among 96 (precision 0.23, recall 0.16, F1 0.19). Random sets of the same size reach F1 0.13 ± 0.00 at their best over the same walk (search-corrected p = 0.0099). 74 genes in that cluster are not on your list.
hunt.tables['candidates'].head(10)| gene_id | product | in_list | distance_to_centre |
|---|---|---|---|
| TGME49_318660 | protein phosphatase PPM5B | False | 0.402 |
| TGME49_271610 | pyrroline-5-carboxylate reductase | False | 1.62 |
| TGME49_267070 | aquaporin 2 | False | 1.66 |
| TGME49_306190 | hypothetical protein | False | 1.72 |
| TGME49_226740 | zinc finger, C3HC4 type (RING finger) domain-containing protein | False | 1.82 |
| TGME49_208550 | hypothetical protein | False | 2.07 |
| TGME49_214950 | hypothetical protein | False | 2.35 |
| TGME49_240570 | hypothetical protein | False | 2.38 |
| TGME49_249200 | Ctr copper transporter family protein | False | 2.42 |
| TGME49_265520 | hypothetical protein | False | 2.46 |
pu = S.run('positive_unlabeled', genes=genes, exclude='compartment')pu.tables['candidates'].head(8)
| gene_id | product | score |
|---|---|---|
| TGME49_253880 | GNS1/SUR4 family protein | 5.11 |
| TGME49_278850 | palmitoyltransferase DHHC2 | 4.56 |
| TGME49_299060 | sodium/hydrogen exchanger NHE2 | 4.46 |
| TGME49_276910 | endoplasmic reticulum lumen protein retaining receptor (ERD2) family protein | 4.45 |
| TGME49_309560 | nmda receptor glutamate-binding chain | 4.43 |
| TGME49_321530 | cathepsin CPL | 4.32 |
| TGME49_250740 | Yip1 domain-containing protein | 4.16 |
| TGME49_301210 | NAD(P) transhydrogenase subunit beta, putative | 4.06 |
profile = S.run('set_enrichment', genes=genes, exclude='compartment')profile.tables['profile'].head(12)
| kind | feature | value | n | effect | p | q |
|---|---|---|---|---|---|---|
| number | n_tm | higher | 135 | 0.355 | 1.17e-102 | 4.3e-100 |
| number | is_tm | higher | 135 | 0.33 | 1.05e-89 | 1.93e-87 |
| layer | struct | denser inside | 78 | 4.41 | 7.54e-78 | 9.28e-76 |
| category | dtm_class | TM | 103 | 2.44 | 2.97e-61 | 2.74e-59 |
| layer | domain | denser inside | 46 | 3.81 | 4.69e-39 | 3.46e-37 |
| number | rna245775_eif12ko_tachy_r1 | higher | 134 | 0.236 | 5.18e-21 | 3.19e-19 |
| number | rna245775_eif12ko_tachy_r2 | higher | 134 | 0.235 | 7.36e-21 | 3.88e-19 |
| number | stress_7-1_unstress_red_24 | higher | 134 | 0.235 | 8.53e-21 | 3.94e-19 |
| number | stress_6-1_unstress_red_24 | higher | 134 | 0.234 | 1.63e-20 | 6.67e-19 |
| number | rna245775_eif12ko_tachy_r3 | higher | 134 | 0.232 | 1.94e-20 | 7.17e-19 |
| number | rna245775_eif12ko_prebrady_r1 | higher | 134 | 0.225 | 2.72e-19 | 9.13e-18 |
| number | rna245775_eif12ko_prebrady_r3 | higher | 134 | 0.222 | 1.05e-18 | 3.24e-17 |
grown = S.run('seed_expansion', genes=genes, exclude='compartment')grown.tables['candidates'].head(8)
| gene_id | product | score | direct_links_to_list |
|---|---|---|---|
| TGME49_257120 | sugar transporter ST1 | 0.000515 | domain 1, orthogroup 1, struct 13 |
| TGME49_288920 | ATP-binding cassette G family transporter ABCG96 | 0.000462 | coexpression 1, domain 2, orthogroup 2, struct 2 |
| TGME49_315560 | ATP-binding cassette G family transporter ABCG77 | 0.000422 | domain 1, orthogroup 2, struct 2 |
| TGME49_229170 | formate/nitrite transporter FNT3 | 0.000415 | cofitness 1, orthogroup 2, struct 2 |
| TGME49_305180 | Na+/H+ exchanger NHE3 | 0.000401 | domain 1, orthogroup 1 |
| TGME49_238160 | hypothetical protein | 0.000388 | orthogroup 1, struct 2 |
| TGME49_299060 | sodium/hydrogen exchanger NHE2 | 0.000387 | cofitness 2, domain 1, orthogroup 1 |
| TGME49_216710 | apicomplexan amino acid transporter ApiAT5-4 | 0.000382 | cofitness 1, domain 7, orthogroup 1, struct 8 |
With twenty or more genes, the self-test of a list strategy hides 30% of YOUR list and asks whether the rest recovers it.
S.test('seed_expansion', genes=genes, exclude='compartment')PASS -- AUROC of hidden members against every other gene 0.824 against 0.495 ± 0.033 under 10 random sets of the same size (bar 0.532, p 0.0909); effect +0.329, needed +0.100; 40 hidden
Hidden: 30% of your list of 135 genes (135 genes). Null: 10 random sets of the same size.
Scorecard (ranking)
| AUROC | 0.824 |
| AUPRC | 0.071 |
| AUPRC lift | 14.180 |
| Prevalence | 0.005 |
| Partial AUROC (FPR <= 10%) | 0.695 |
| R-precision | 0.100 |
| Precision @ top 1% | 0.148 |
| Enrichment @ top 1% | 29.796 |
| Recall @ top 10% | 0.525 |
| Best F1 | 0.211 |
| nDCG | 0.477 |