Gene network inference challenge

Organized by Goudet - Current server time: Feb. 24, 2020, 7:43 p.m. UTC


Training phase
Jan. 22, 2020, midnight UTC


Test phase
Feb. 4, 2020, midnight UTC


Competition Ends

The goal of this challenge is to reverse enginner gene regulation networks from simulated steady-state observational data. Participants are challenged to infer five network structures of 20 or 100 nodes.

Evaluation Criteria

Submission : during the training and test phases, each participant is challenged to infer 5 networks of 20 or 100 nodes. The particpants have to provide a zip folder containing 5 .csv files (prediction_network1.csv,..., prediction_network5.csv) corresponding to its predictions for the 5 networks. In a prediction file for a given network each line must correspond to the prediction score between 0 and 1 for each directed edge from gene i (cause) to gene j (effect), where gene i and gene j are two different genes of the dataset (self-loop are excluded). Each prediction score corresponds to the confidence of the participant in the existence of the directed edge. If a link is missing in the prediction file, it will be assumed that the prediction score is equal to 0. An example of prediction file for the network1 of the training dataset is provided in the starting kit package.

Evaluation : in the evaluation process, for each network we count as a true positive and edge from gene i to gene j of the true network which is correctly recovered by the participant. Tp is the number of true positive. A false negative is a directed edge not in the predicted network which is in the true network. Fn is the number of flase negatives. A false positive is an edge in the inferred network which is not in the true network (reversed edges and edges which are not in the skeleton of the true network). Fp is the number of false positives. For this challenge we use an evalaution with the precision-recall curve showing the tradeoff between precision (Tp/(Tp+Fp)) and recall (Tp/Tp+Fn)) for different causation thresholds. This precision-recall curve can be summarized by the Area under the Precision Recall Curve (AUPR), ranging from 0 to 1, with 1 being the best score. The goal of this challenge will be then to obtain the best average AUPR score over the 5 networks (in the test phase).

Training phase

Start: Jan. 22, 2020, midnight

Test phase

Start: Feb. 4, 2020, midnight

Competition Ends


You must be logged in to participate in competitions.

Sign In
# Username Score
1 rayan 0.3124
2 taibeche 0.3016
3 aabbas 0.2954