On 2014 Jun 05, Robert M Flight commented:

Unless I misread the paper, the NMAD that is used for assigning weights is derived from the exact same data that subsequently undergoes biclustering. A better approach might be to derive weights from independent experiments and apply those to the data that one wishes to bicluster.

In addition, the difference in significance with and without (alpha=0) the incorporation of the weights seems to imply that the use of the weights adds very little information to the biclustering algorithm, and makes one question as to why they should be calculated at all. The biclustering algorithm using the networks shows a definite difference compared to the other methods tested, but the weighted vs unweighted biclustering shows very little differences, making one question the significance of the "weighted co-clustering approach", vs a "network based co-clustering approach".

On 2014 Jun 05, Robert M Flight commented:

Unless I misread the paper, the NMAD that is used for assigning weights is derived from the exact same data that subsequently undergoes biclustering. A better approach might be to derive weights from independent experiments and apply those to the data that one wishes to bicluster.

In addition, the difference in significance with and without (alpha=0) the incorporation of the weights seems to imply that the use of the weights adds very little information to the biclustering algorithm, and makes one question as to why they should be calculated at all. The biclustering algorithm using the networks shows a definite difference compared to the other methods tested, but the weighted vs unweighted biclustering shows very little differences, making one question the significance of the "weighted co-clustering approach", vs a "network based co-clustering approach".