Sign In

ASAP Discovery x OpenADMET CompetitionTake part in the first prospective benchmark on Polaris.

Benchmark

polaris/pkis1-egfr-wt-mut-c-1

A multitask classification benchmark for kinase EGFR wild type and L858R mutant.

Created on: December 08, 2023Train size: 290Test size: 74
Public
Multi Task
Classification

Participants

P

Tags

kinase
hit-discovery
selectivity

Related dataset

polaris/kinase-pkis1-1
2023-12-08

Leaderboard

Test set
Task
#
NameContributors
accuracy
f1
roc_auc
pr_auc
mcc
cohen_kappa
References
1

pkis1-egfr-wt-mut-c-1_DGLGraphTransformer_GCNModel

0.365
0.356
0.615
0.217
0.223
0.095
No references provided
2

pkis1-egfr-wt-mut-c-1_DGLGraphTransformer_MPNNModel

0.176
0.299
0.500
0.176
0.000
0.000
No references provided
3

pkis1-egfr-wt-mut-c-1_DGLGraphTransformer_GATModel

0.176
0.299
0.500
0.176
0.000
0.000
No references provided
4

pkis1-egfr-wt-mut-c-1_secfp_FCModel

0.176
0.299
0.500
0.176
0.000
0.000
No references provided
5

pkis1-egfr-wt-mut-c-1_DGLGraphTransformer_AttentiveFPModel

0.176
0.299
0.500
0.176
0.000
0.000
No references provided

Details

README

molprop

Background

EGFR (Epidermal Growth Factor Receptor) kinase is a type of receptor tyrosine kinase that plays a significant role in cell growth, proliferation, and survival. Mutations or overexpression of EGFR have been associated with various diseases, particularly cancer.

Benchmarking

EGFR Wild type: Targeting wild-type EGFR with small-molecule inhibitors, such as erlotinib, is an ongoing area of research in the treatment of glioblastoma. While early findings are promising, the complexity of glioblastoma biology presents challenges that require further investigation to improve treatment outcomes for patients.

EGFR L858R: While EGFR TKIs initially demonstrate impressive responses in NSCLC patients with the L858R mutation, resistance to these drugs can develop over time. However, newer generations of EGFR TKIs, like osimertinib, have been developed to target these resistant mutations.

The goal of this benchmark is to select the best predictive model for

  • Optimization of the bioactivity % inhibition.
  • Discovery of potential hits in new chemical space.

Description of readout

  • Readouts: CLASS_EGFR, CLASS_EGFR_(L858R_mutant)
  • Bioassay readout: percentage of inhnibition.
  • Optimization objective: postive label (1)
  • Number of data points: train: 290 test: 74
  • Thresholds:
    • EGFR_(L858R_mutant) > 75%
    • EGFR > 75%

Data resource:

Train/test split

Given the benchmarking goal, a scaffold-based splitting approach was applied to ensure training and test sets contain distinct chemical structures while maintaining the diversity of scaffolds.

Distribution of the train/test in the chemical space image

Related links

The full curation and creation process is documented -> notebook

Related benchmarks

  • polaris/drewry_egfr_wildtype_singletask_reg_v1
  • polaris/drewry_egfr_wildtype_singletask_clf_v1
  • polaris/drewry_egfr_wt_l858r_multitask_reg_v1

Note: It's recommanded to evaluate your methods agaisnt all the benchmarks related to this dataset.