PubMed

A typographical error occurred in the title. The correct title would start with "Laurate Permeabilizes..."

Rationale of the method applied in Arnedo et. al.

Clustering methods have been applied for long time in pattern recognition problems and in a variety of fields and actually are embedded in most of our devices, from a simple refrigerator to a Boing engine. In biomedical sciences there is a trend of considering that few classes of clustering exist (e.g., hierarchical cluster of K-Means). However, datamining, knowledge discovering, and machine leaning are fields of computer sciences with very rich and different either theoretical or empirical methods. Much of the power of these methods is based on the fusion of them, taking advantages of each one.

The methodology and framework proposed in this Arnedo et. al encodes a generalized factorization method (GFM) that was designed to identify structural patterns or clusters (substructures) that characterize complex objects (1-8) embedded in databases (6, 7, 9, 10). Unfortunately, solving these kinds of complex problems cannot be performed by a single clustering method but utilizing and combining the advantages of many of them, as were implemented in the GFM and summarized below.

First, we utilized the notion of Model-based clusters (5, 11, 12), where the centroid/prototype of a cluster is not a point but a model (e.g., line segment, ellipsoids, (5, 13, 14) or see C-varieties algorithm (12)) -or a family of models (8) - defined as a relation between cohesive subset of points that meet certain constraints (e.g., Relational clustering (12), undirected graph (12)). When the subjacent equations of the models are unknown, they can be estimated from the data by identifying relations between subsets of observations that show similar patterns under a specific subset of relevant descriptive features (attributes) (4, 5, 15). Here we proposed that in a pure data driven analysis these unknown models (grey boxes) can be learned as local relationships between subsets of features and observations, which are termed biclusters. The biclusters are often represented as sub-matrices (4, 5, 15). Other models proposed in this work (black boxes) consist of those that are encoded by a particular method devoted to recognize certain classes of patterns.

Second, to identify a family of models (8) or biclusters we planned the use of the NMF (16) factorization algorithms (tensor analysis), which can uncover cohesive data represented as sub-matrices (factors). When sorted and thresholded, each sub-matrix represents a bicluster. Notably, biclustering produces several local models of the data (17), each one capturing intrinsic relationships between subsets of observations sharing subsets –instead of all- features. Diverse biclusters –instead of uniform data explanations- provide a better understanding of the described object that is diluted when together, all features are used in a single global model of data typically performed by clustering methods (17).

Third, we proposed in this work that biclusters can be fuzzy (overlapping, see Fuzzy clustering (18)), where voxels and/or subjects can belong to more than one bicluster), as well as possibilistic (non-exhaustive partitions, see Possibilistic clustering (12)), where certain voxels and/or subjects may not be assigned to any bicluster, and thus, outliers can be detected.

Fourth, we demonstrated that biclusters could be identified without choosing a priori a particular number of clusters (11, 13, 14, 19). Setting this parameter as a small number may eventually generate few but large data partitions here defined as general biclusters that may underfit the data. In contrast, using a big number of clusters will generate many clusters with partitions of small size (i.e., more granular partitions) here defined as specific biclusters that may overfit the data (1, 5, 11, 13, 14, 19-21). Although there are many different validity indices (12, 18) that suggest the best number of clusters for a given dataset, they often produce contradictory results (11, 13, 14, 19). This problem is exacerbated when Centroid-based (e.g., k-means, NMF) or Distribution-based clustering (e.g., EM) is utilized because two successive numbers of clusters may generate completely different cluster arrangements (see (12, 18) for a review). Instead of fighting over the lost battle of uncovering a single optimal number of clusters, biclusters were calculated from partitions generated by all number of clusters between 2 and √n (12), where n is the number of observations (subjects). We postponed the selection of the best clusters until all partitions were examined to select a set of clusters that together provide an optimal description of the sample. These clusters could be chosen from different partitions that were generated by different number of clusters (see Consensus clustering (1, 5, 11, 13, 14, 19-21)).

Fifth, formulation of the bicluster identification problem from different partitions may produce redundant, excessively specific and/or excessively general biclusters (4, 5, 15, 22, 23). To select optimal biclusters, we proposed a GFM that performs the following Consensus clustering strategy: (i) eliminates all redundant biclusters by comparing the similarity between their subjacent sub-matrices using Hypergeometric statistics (see Methods and Material in Supplement 1), and selecting only one representative bicluster from each set of equivalent sub-matrices. Biclusters harboring <10% of the total number of subjects were not considered to avoid a trend to obtain singleton biclusters. (ii) From the non-redundant biclusters, GMF selects all biclusters that are optimal in terms of specificity, generality and diversity. To do so, it selects all non-dominated biclusters where one bicluster is not worse (i.e. dominated) than another in both objectives: specificity and generality (4, 5, 15, 22, 23). Moreover, to ensure diversity, GMF applies the non-dominance metric only among biclusters within a neighborhood (22, 23) (i.e., biclusters with a relatively high overlapping of subjects and voxels with). This Multiobjective and Multimodal optimization process is analogous to Minimum Description-Length methods (24) and was carried out as described in (1-8). (iii) The remaining optimal biclusters are hierarchically organized by inclusion of their subjects into connected or disjoint hierarchies (i.e., subgraphs or subnetworks).

Sixth, and after the first step of identifying interesting patterns describing objects embedded in the data, we went one step further by also finding characteristic descriptions for each group, where each group represents a concept or class using Conceptual clustering. In this work we incorporated factors such as the generality and simplicity of the derived bicluster descriptions by (1) relational clustering validation against data driven independent descriptions obtained in other domains of knowledge, and (2), incorporate the previous structured knowledge into predictive multiclassifiers mixing machine learning and multiobjetive optimization techniques.

Dear Authors,

We have published an analysis in S. cerevisiae, showing expression-dependent artifactual ChIP enrichment at highly expressed loci (Teytelman L, 2013 "Highly expressed loci are vulnerable to misleading ChIP localization of multiple unrelated proteins"). As you know, our finding raises the question of whether HOT regions may also be influenced by the same artifact.

It is great that you have considered our work and have thoughtfully responded to our analysis. Below, I would like to continue this discussion in an effort to better understand the artifact, its causes, and whether it may be contributing to the enrichment at the HOT loci.

1. “we have demonstrated that there is no correlation between our non-specific binding controls (IgG) and our measured transcription factor occupancy;”

Considering our results with no-tag control experiments, an IgG may fail to control for the artifact. It would be great if you could instead perform a GFP ChIP-Seq, similarly to what we have done in yeast.

2. The regions determined in ref. 41 have very low enrichment (twofold or less) of non-specific immunoprecipation in anti-GFP antibody controls over input DNA evaluated using a non-standard sliding-window approach. Importantly, immunoprecipitation/input ratios at this level are typically not considered enriched for binding in modern peak-calling procedures. For example, the median immunoprecipitation/input ratio for our human RNA Pol II experiments is 20-fold, and only 0.033% of human RNA Pol II peaks contain an immunoprecipitation/input ratio ≤ twofold.

The mean is low, but in both anti-GFP experiments, there are loci with 3-5x enrichment (figure 4D). Most importantly, while the anti-GFP enrichment at the hyper-ChIPable loci is low, please note that the level of enrichment is variable from protein to protein (2-5X for Sir proteins, but often >10X for Cse4).

3. Thus, it is essential to note that the term ‘hyper-ChIPable’, coined by ref. 41, is quite misleading, as a correctly performed ChIP experiment will evaluate statistically enriched regions, with higher immunoprecipitation/input ratios. The so-called hyper-ChIPable regions in ref. 41 are not binding regions as determined under ChIP-seq best practices. Hence, when statistical peak-calling was performed in ref. 41 (using the established MACS peak-caller) to evaluate signals only at significantly enriched regions (Supplementary Table 1) only 17 (<7.5%) of the 238 claimed ‘hyper-ChIPable’ regions were called significant by all three Sir proteins. In fact, 68% of their 238 regions do not contain a binding site for any Sir protein as determined by MACS, despite even very liberal settings used (P < 10−5, no fold enrichment cut-off). Thus, the data of ref. 41 contradict its own major claim that all three Sir proteins showed enrichment at the 238 sites.

By reporting the 238 sites with >2fold enrichment of Sir2, Sir3, and Sir4, we are in fact being extra-demanding in terms of the threshold. We are stringently requiring all three proteins to be enriched above a threshold at the locus. So a target with 5x enrichment of Sir2 and 1.8X enrichment of Sir3 would not pass this cutoff. A typical ChIP study will focus on a single factor at a time. Had we done that, we would have many more artifactual targets for each silencing protein, with many at 5x or higher enrichment. Furthermore, the level of the artifactual signal varies from protein to protein or experiment to experiment. For example, the Cse4 signal at highly-expressed loci can give 10x or higher enrichment.

4. Furthermore, as indicated in Supplementary Table 3 of ref. 41, the Sir2, Sir3 and Sir4 ChIP-seq experiments were performed only once each, which raises the question as to whether enrichment of Sir proteins at the 238 sites is reproducible. More rigorously, even for the remaining 17 genomic loci, their status as hyper-ChIPable is questionable as each region would first have to be established as a reproducible binding site in replicate experiments for each individual Sir protein. If you consider that Sir2, Sir3 and Sir4 ChIP-seq constitutes three replicates of Sir proteins, their data show that most of their claimed sites were not reproducibly enriched.

Most of our artifact-cause analysis focuses on genome-wide data, not on the 238 sites. The 238 Sir-enriched euchromatic loci were a launching point for the analysis, but most of the paper looks comprehensively at the link between expression and ChIP levels. Figures 3, 4, and 5 are all on genome-wide correlations between Pol II/III and ChIP.

As for reproducibility, we see the same peaks, with often 10x enrichment, in Ste12, Cse4, two distinct GFP experiments, and each of the three Sir ChIP-Seq datasets. The same exact loci come up in the Sir3 paper from Oliver Rando’s group (Radman-Livaja M, 2011).

5. In addition to the analytical differences outlined above, other potential sources for the marked differences between our data and the Sir-enriched regions of ref. 41 are deviations from a typical ChIP protocol. In particular, ref. 41 employed a significantly longer cross-link time (1 h as opposed to the typical 10–20 min). This might contribute to formation of large non-specific protein–DNA complexes, which can in turn increase non-specific immunoprecipitation.

Though not discussed in the manuscript, we have in fact performed experiments to investigate if the crosslinking concentration contributed to the misleading signal. We performed ChIP with the 1 hour crosslinking at room temperature at the following formaldehyde concentrations: 0.0625%, .125%, .25%, .5% and 1%, but did not find a proportionate decrease in the hyper ChIPpable signal with the decreasing formaldehyde concentrations. Moreover, the presence of hyper-ChIPability in the Snyder datasets (Cse4, Ste12), ours (Sir2, 3, 4, GFP), and Rando (Sir3) make it clear that the problem is not in some unusual protocol steps in our hands.

We also note that we initially performed the Sir ChIP-Seq experiments because of our interest in the Sir protein biology. Because the Sir proteins do not directly interact with the DNA, we used longer crosslinking times. This is not unique to our work.

In summary, much more work is needed to pinpoint the cause of the artifact and to evaluate whether some or all of the signal at highly expressed genes in many other reported ChIP studies could be artifactual. Much more work is necessary to develop the best controls and corrections for the artifact. However, the artifact we report is not minor and is not a consequence of the methodological details of our manuscript.

Also, please note the following papers, published almost in parallel with ours, on this topic:

Park D, 2013 "Widespread Misinterpretable ChIP-seq Bias in Yeast" (Different analysis methods but the same conclusions in S. cerevisiae, analyzing an entirely different set of factors with ChIP-Seq experiments.)

Kasinathan S, 2014 "High-resolution mapping of transcription factor binding sites on native chromatin" (Questions specificity of standard ChIP in S. cerevisiae and at HOT regions of Drosophila. This work possibly provides a solution to the artifact with a modification of the ChIP technique.)

Also, the following discussion of our work on PubPeer may be useful.

Wonderful!. This is a must read for all diagnosticians. But don´t miss Stanley DE, Campos DG. The logic of medical diagnosis. Perspect Biol Med. 2013 Spring;56(2):300-15 that takes the argument one step further.

Yes, the abductive mode of thinking that Peirce described is really very crucial in all forms of diagnostics. Depending on the problem at hand we clearly use different logical strategies. This is a must read, together with Lawson AE1, Daniel ES. Inferences of clinical diagnostic reasoning and diagnostic error.J Biomed Inform. 2011 Jun;44(3):402-12. Sadly this sort of analysis seldom appears in clinical speciality journals. If you are lucky you can find articles on Dual System theory and Bayesian theory, but without abduction there is no place to start using these techniques. We all need a dose of metacognitive musings from time to time.

False conclusions drawn about the level of evidence of factual statements drawn from Wikipedia

Paul Vaucher, PhD, DiO^1, Jean Gabriel Jeannot, MD^2, Reto Auer, MD, MAS²

1 University Center of Legal Medicine Lausanne-Geneva, University Hospital of Lausanne (CHUV), Lausanne, Switzerland

2 Department of Ambulatory Care and Community Medicine, University of Lausanne, Lausanne, Switzerland

We believe Hastly et al's study (Hasty RT, 2014) to be misleading and that their paper should not be made available to the scientific community without serious revision. Apparently, it has passed unnoticed that the methodology and statistical analysis they used had little to do with their stated hypothesis and that their interpretation of the statistics was erroneous. Authors concluded Wikipedia to be an unreliable source of health information when an alternate interpretation of the presented results would in contrary point towards considering Wikipedia as a trusted source of information, provided rigorous reanalysis and reinterpretation. There conclusions were therefore in contradiction with other studies on the subject (Archambault PM, 2013, Kräenbring J, 2014) that tend to show that for topics for which health workers contribute, such as for drugs, Wikipedia’s information as trustworthy as those from textbooks. These discrepancies can be explained by major statistical and methodological errors in Hastly et al’s publication.

The correct interpretation of the McNemar statistic suggests that the concordance for diabetes and back pain are significantly better than for concussion, and not the reverse, as stated by the authors. Using data provided in Table 3, for factual statements identified by both reviewers (two first columns of Table 3), for diabetes mellitus and back pain, the authors found that up to 94% of assertions on Wikipedia were verified (respectively 72 out of 75 statement and 63 out of 67, p<0.001 for NcNemar statistic). In contrast, for concussion, only 65% were verified (66 out of 98, p=0.888 for NcNemar statistic ). The McNemar statistic tests whether the proportion of factual statements from both reviewers, classified as concordant or discordant by the authors, are above what would be expected by chance alone (i.e. 50%). The interpretation for diabetes mellitus, if McNemar’s test should be used at all, is that we would fail to reject the null hypothesis of a proportion of concordance of 50% in < 1% of the cases. McNemar statistic thus suggests that the concordance for diabetes and back pain are significantly better than for concussion, and not the reverse, as stated by the authors. The calculation is based on the number of discordant results on the diagonals (i.e. 34 vs 1 for diabetes mellitus). It does by no means test the discordance between Wikipedia statements and existing guidelines. To test their hypothesis, it would appear more relevant to simply report the pooled average proportion of correct statements with their confidence intervals. On a technical note, McNemar statistic should not be used when there are different numbers of assertions assessed between reviewers. McNemar is also known to be highly dependent on the number of factual statements within each article. Article with higher number of statements would reach the level of significance with higher proportion of ungrounded factual statements.

Second, the article falsely leads readers to believe that all factual statements (assertions) from 10 Wikipedia articles were identified and independently assessed by two internist to see whether they were concordant or not. Using two reviewers is a recognized method for increasing precision of a measure. However, the authors did not to provide statistics allowing readers to assess the between-reviewers variability in the identification of assertions. Table 3 suggests that over a third of assertions were reported by only one of the two reviewers. Authors did not find a method to then resolve these dissimilarities (to use their definition in Table 2) and clearly define which assertions were factual statements and which were not. They then did not to define a method of agreement to define which assertions were supported by evidence and which were not. Analysed results therefore only tend to show that for certain topics, internists have difficulties in detecting factual statements from Wikipedia and knowing whether they are grounded or not.

Hastly et al’s findings suggest that while there might be some discrepancies in the quality of articles between topics, some appear of very high quality, such as diabetes mellitus and back pain. Given the unnoticed errors included in their article and the importance on the interpretation of the results, Hastly et al.’s published article reveals that peer reviewed misleading information can also be made available to the public.

Conflicts of interest : Reto Auer and Jean-Gabriel Jeannot are advocates of the use of Wikipedia as a communication mean to inform the population on health issues. Paul Vaucher is an important contributor to the French Wikipedia page dedicated to osteopathic medicine.

A good overview, but would be so much more useful if the authors could list the 1,453 (figshare?). Even better if they surfaced a resolution of these to molecular entites via PubChem CIDs. Note they could also submit not only this set but the previous seven therapeutic splits and get more traffic for the set of papers.

Post-publication commentary:

"ARISE, and Cast Off the Shackles of EGDT"

The sound you hear is a sigh of relief from Emergency Physicians and intensivists regarding the outcomes of the Australasian Resuscitation in Sepsis Evaluation (ARISE).

As ProCESS suggested, and as many have suspected all along, it seemed the critical intervention from Early Goal-Directed Therapy was the early part – and less the SCO2 monitoring and active management of physiologic parameters using dobutamine and blood transfusion. Now, we have a second study, in addition to ProCESS, supporting the same general conclusions....

http://www.emlitofnote.com/2014/10/arise-and-cast-off-shackles-of-egdt.html

In the section on guidelines, the authors have stated that the current British Committee for Standards in Haematology guidelines Davies JM, 2011 recommend lifelong penicillin prophylaxis for all persons with surgical asplenia. Although this was true in previous guidelines Working Party of the British Committee for Standards in Haematology Clinical Haematology Task Force, 1996, Davies et al recommend a risk adapted approach to antibiotic prophylaxis. They recommend antibiotic prophylaxis "at least" in the immediate post-operative period following splenectomy for trauma. Otherwise, long term prophylaxis is recommended for those considered to be high risk of pneumococcal infection: age <16 years or >50 years, inadequate serological response to pneumococcal vaccination, previous invasive pneumococcal disease and splenectomy for underlying haematological malignancy.

Additionally, the authors do not discuss the meningitis serogroup B vaccine (MenB). A MenB vaccine was licensed by the European Medicines Agency in January 2013. Although not yet FDA licensed, it has been used to control several outbreaks in the United States http://www.cdc.gov/meningococcal/outbreaks/vaccine-serogroupb.html. In England, over 80% of Neisseria meningitidis cases are due to capsular group B strains and it has been recommended in England for children and adults with asplenia, splenic dysfunction and complement disorders https://www.gov.uk/government/collections/immunisation-against-infectious-disease-the-green-book.

To finalize three-dimensional VDAC structure: Focus on Native VDAC will be indispensable

This study, from my point of view, marks a great moment of VDAC research:

1) It points another time to the relevance of cell membrane-standing VDAC-1 for the pathogenesis of Alzheimer´s Disease via apoptosis.

2) It gives strong support to the cell membrane-expression, more precisely plasmalemmal lipid raft-integration of vertebrate VDAC-1. Furthermore, the data concerning the posphorylation of VDAC-1 in correlation to regulation of channel opening/closing argue in favor of its involvement in cell volume regulation and thus apoptosis. They are in line with much evidence indicating that plasmalemmal VDAC-1 forms the channel part of a volume regulated anion channel complex (VRAC/VSOAC).

3) From here, VDAC-1 in the plasmalemma must be “fully closed” = collapsed = N-terminus accessible outside the barrel = closed for anions and cations, and there is evidence that the N-terminal part of native VDAC-1 can be reached by antibodies even in detergent solutions (Benz et al., 1992; Thinnes and Burckhardt, 2012).

4) Applying canonical incorporation into black membranes, detergent-solubilized native phosphorylated mammalian VDAC-1 as well as recombinant channel preparations from E. coli inclusion bodies show only “open” = anion-selective = N-terminal stretch inside the barrel and “closed” = cation-selective = semi-collapsed = N-terminal stretch inside the barrel channel phenotypes (Teijido et al., 2012). “Fully closed” = collapsed = N-terminus accessible outside the barrel VDAC-1 states thus cannot be studied by this approach. The same holds true for more recent crystallization-based approaches. However, upcoming laser-based approaches may work just on native VDAC-1 in solutions; thus improvements to get detergent solubilized native VDAC preparation may pay to keep on the schedule.

--Benz R, Maier E, Thinnes FP, Götz H, Hilschmann N (1992) Studies on human porin. VII. The channel properties of the human B-lymphocyte membrane-derived “Porin 31HL” are similar to those of mitochondrial porins, Biol. Chem. Hoppe Seyler. 373: 295–303. --Thinnes FP, Burckhardt G (2012) On a fully closed state of native human type-1 VDAC enriched in Nonidet P40. Mol Genet Metab 107: 632-633. --Teijido O, Ujwal R, Hillerdal CO, Kullman L, Rostovtseva TK, Abramson J (2012) Affixing N-terminal α-helix to the wall of the voltage-dependent anion channel does not prevent its voltage gating. J Biol Chem 287: 11437-11445.

The authors predict that 36% of bacteria cannot synthesize phenylalanine, but many (most?) of these predictions are erroneous. I looked over the predictions, which were graciously provided by Dr. Kost, and noticed the following errors. First, many cyanobacteria are predicted auxotrophs (i.e., Synechococcus elongatus PCC 6301, and members of the genera Prochlorococcus, Anabaena, Cyanothece, Nostoc, and Gloeobacter) . However cyanobacteria are normally grown in a mineral medium with no organic carbon. Second, our group studies the sulfate-reducing bacteria Desulfovibrio alaskensis G20 and D. vulgaris Miyazaki F, which grow in a defined minimal medium without any amino acids, yet these are predicted to be phenylalanine auxotrophs. Third, Caulobacter crescentus CB15 is predicted to be a phenylalanine auxotroph, but it can grow at very low nutrient levels, including in defined mineral media with a small amount of sugar added.

PubMed soon should be adding the published Erratum to this Letter to the Editor (responding to academics' research important to risk assessment (Krauth D, 2013), but as I prompted the Erratum, I wish to add that it, concerning the declared interests (DoI) of two of these industry and industry-affiliated authors (Pr.'s Leist & Boobis), even now fails to say 'industry' or a synonym; whereas my investigations revealed that these two also work regularly for industries.