Skip to main content
Premium Trial:

Request an Annual Quote

Like a Tennis Match

In a series of blog posts at Bits of DNA, computational biologist Lior Pachter criticizes a pair of papers appearing in Nature Biotechnology over the summer presenting methods to interrogate biological networks. Pachter in particular has harsh words for the Massachusetts Institute of Technology's Manolis Kellis. Pachter accuses Kellis's recent paper of being "dishonest and fraudulent." Kellis, in turn, says the accusations are "defamatory, baseless, misleading, and factually incorrect." A bit of a back-and-forth ensues.

The first paper Pachter critiques, from Baruch Barzel and Albert-László Barabási, presents a method to silence indirect effects within a network by drawing on dynamical correlations within that network. According to Pachter, the approach detailed in the paper does not actually show a statistically significant improvement over other approaches — he calls the validations "shoddy" — and is of limited "current practical utility."

The paper from Kellis, meanwhile, describes a network deconvolution method to infer direct effects from an observed correlation matrix containing both direct and indirect effects. Pachter argues that the method used to get results differs from what the authors presented in the main text of the paper, that the way to set parameters is not described, and that the results are not reproducible, among others. Pachter also writes that a clarification made to the supplementary material section of the paper appears to include a changed, rather than clarified, figure.

Kellis and his group respond to Pachter both on Pachter's blog in the comments and in a separate document. "We stand firmly by the scientific contributions of our work to the field of network science, and address the six allegations [Pachter made]," the group writes. They then address those allegations point by point, saying that much of the information Pachter seeks is included within the paper, supplementary materials, or figures. They also say their lab has provided their code to a number of researchers who were able to implement it or reproduce their findings. Kellis and his colleagues acknowledge that the correction made to a supplementary figure was not fully documented, but say that that omission does not affect their conclusions and that the correction notice has been updated.

They also suggest that the close inspection of their work is "personally motivated." Pachter, in a third post, recounts reviewing a paper from Kellis's group in 2006 and giving it a negative review, saying that the algorithm did not "achieve its stated goals," though the paper was ultimately accepted to the conference. Again, Kellis's team "stand[s] by the novelty and contribution of our work" and says that it has been "replicated and built upon," here.

Pachter says that the response regarding the Nature Biotechnology paper "falls short," adding that "[t]he distance between the image created by the main text of the paper and the truth of their method is simply too great."

In the comments, Kellis says that his team "continue[s] to disagree, and we stand firmly by the specific presentation of our work, with a broadly accessible main text that consistently refers to a much more formal and detailed supplement."