search

dkoslicki / NCATS

MIT License
2 stars 0 forks source link

Random Walk with Restart development #98

Closed saramsey closed 6 years ago

saramsey commented 6 years ago

This is a placeholder issue for the development of Random Walk with Restart for multi-path relatedness between a "Q1 disease" and a genetic condition (OMIM).

saramsey commented 6 years ago

Update from @chyods:

Dear prof. Ramsey,

Here is the update to RWR with edge-weight training. RWR and weighted RWR shows the ranked position of the correct answer for each query returned by RWR and weighted RWR, respectively. The smaller the value the better the result is.

(see table in next comment)

Overall, the weights for edges disease_affects, gene_assoc_with and phenotype_assoc_with have to be increased by roughly 100x, 60x and 40x. I'm still unsure how to improve the last query (Ashtma), but I'm still working to see if I could improve this one while minimizing the loss of accuracy of other queries.

Please let me know if you have any question or suggestion regarding this results.

Thank you very much, Yodsawalai

saramsey commented 6 years ago
query MeSH RWR MRR weighted RWR MRR #candid MRR diff
DOID:11476 Osteoporosis 3122 0.0003 76 0.0132 17752 40.08
DOID:1498 Cholera 26 0.0385 24 0.0417 8482 0.08
DOID:12365 Malaria 45 0.0222 39 0.0256 8482 0.15
DOID:13810 Hypercholesterolemia 9 0.1111 3 0.3333 17752 2.00
DOID:9352 Diabetes Mellitus, Type 2 2214 0.0005 77 0.0130 17752 27.75
DOID:2841 Asthma 305 0.0033 682 0.0015 8482 -0.55
dkoslicki commented 6 years ago

This issue was moved to RTXteam/RTX#11