11 major claims checked against the paper's own evidence: 1 not fully backed by the presented evidence (unsupported or overstated).
unsupportedDiscussion, final paragraphReviewer 2
SCiFI can be applied beyond microbiomes, e.g., to neural firing rates and T-cell receptor sequences.
The paper suggests these applications in the Discussion but provides no experiments or analysis to support them. This is a speculation, not a demonstrated claim.
Evidence: None; only a forward-looking statement in the Discussion.
“As a first example, SCiFI may be applied to neural firing rates to learn the low-dimensional neural manifold underlying motor functions.”
partialAbstractReviewer 2
The integrated approach yields simple structure-function maps of microbiomes, allowing the discovery of molecular mechanisms underlying human and environmental health.
The paper shows that SCiFI yields simple maps and reveals a mechanistic hypothesis for pH sensitivity in soil denitrification. However, the claim extends to 'human health' without direct human health data; the gut and ocean examples are at the mechanistic level but do not directly address health outcomes.
Evidence: The paper presents a mechanistic model for soil denitrification and references health relevance for the gut (butyrate) but does not test health outcomes.
“This integrated approach yields simple structure-function maps of microbiomes, allowing the discovery of molecular mechanisms underlying human and environmental health.”
supportedResults - SCiFI algorithm correctly identifies functional groups for in vitro gut microbiomesReviewer 1
SCiFI correctly identifies functional groups in synthetic gut microbiomes.
The paper demonstrates that SCiFI recovers known functional groups (e.g., butyrate producers, pH buffers) in a synthetic gut community, with validation via comparison to experimental phenotypes from Clark et al. (2021).
Evidence: Figures 3a-d show that SCiFI achieves high predictive accuracy (R² near 0.9) and the groups are biologically interpretable.
“SCiFI algorithm correctly identifies functional groups for in vitro gut microbiomes”
supportedResults - SCiFI finds functional groups in natural ocean and soil microbiomes.Reviewer 1
SCiFI finds functional groups in natural ocean and soil microbiomes.
The paper shows that SCiFI identifies two groups for ocean nitrate prediction and three groups for soil nitrate dynamics, with consistency across test-train splits.
Evidence: Figures 3i-p show predictive performance and group structure for ocean and soil datasets.
“SCiFI finds functional groups in natural ocean and soil microbiomes.”
supportedResults - SCiFI distills groups of genes in the ocean metagenomeReviewer 1
SCiFI distills ~500 gene modules into three sparse groups highlighting survival strategies in the ocean metagenome.
The paper presents results showing that three groups suffice to predict environmental variables, and the sparsified groups contain interpretable gene modules related to survival strategies (e.g., degradation, protective compounds).
Evidence: Figure 4 and accompanying text describe the three groups and their correlation with depth and environmental variables.
“Three groups suffice to make accurate predictions (; ).”
supportedResults - SCiFI learns functional groups that are dynamical variables of mathematical modelsReviewer 1
The functional groups learned by SCiFI can be directly used as variables in a mathematical model (consumer-resource model) that predicts community function.
The paper demonstrates that a two-group consumer-resource model, parameterized by SCiFI-identified groups, accurately predicts nitrate dynamics and biomass across different pH levels.
Evidence: Figure 5 and associated text show model predictions matching data, with group compositions identified by SCiFI.
“SCiFI learns functional groups that are dynamical variables of mathematical models”
supportedResults - SCiFI guides targeted experiments that reveal biological mechanismsReviewer 1
SCiFI guides targeted experiments that reveal biological mechanisms (e.g., genetic basis for differential pH sensitivity in soil denitrifiers).
The paper shows that genome sequencing of isolates from the two groups reveals distinct denitrification gene sets, and these differences explain community pH sensitivity.
Evidence: Figure 6 and the 'SCiFI guides targeted experiments that reveal biological mechanisms' section.
“SCiFI guides targeted experiments that reveal biological mechanisms”
supportedResults, 'SCiFI algorithm correctly identifies functional groups for in vitro gut microbiomes'Reviewer 2
SCiFI correctly recovers known functional groups in gut microbiomes.
The paper shows that SCiFI identifies the same groups (e.g., pH buffers, butyrate producers) as independently confirmed in a prior study.
Evidence: Figure 3a-d shows that SCiFI's groups align with the expected biology: e.g., Anaerostipes caccae as a key butyrate producer.
“These groups, which were independently experimentally confirmed in Ref. [] show that SCiFI correctly learns functional groups in the microbiome.”
supportedResults, 'SCiFI distills groups of genes in the ocean metagenome'Reviewer 2
SCiFI distills ~500 gene modules down to three sparse groups in the ocean metagenome.
The paper shows that SCiFI, with gating, reduces the genes to three interpretable groups correlated with environmental variables.
Evidence: Figure 4 shows three groups with correlated abundances to nitrate, temperature, and oxygen, and highlights relevant KEGG pathways.
“Three groups suffice to make accurate predictions (; ). Two groups are strongly correlated to nitrate concentration (blue; group 1) and temperature (red; group 3)”
supportedResults, 'SCiFI learns functional groups that are dynamical variables of mathematical models'Reviewer 2
SCiFI identifies two functional groups in soil that enter a mathematical model of nitrate metabolism.
The paper demonstrates that two groups found by SCiFI can be used in a consumer-resource model that accurately predicts nitrate dynamics and is robust to randomization tests.
Evidence: Figure 5 shows model predictions matching data; null model comparison shows randomized groupings degrade performance.
“We find that, with gating, two groups suffice to predict nitrate dynamics across the entire dataset”
supportedResults, 'SCiFI guides targeted experiments that reveal biological mechanisms'Reviewer 2
SCiFI-guided targeted experiments reveal that denitrification differences between groups are explained by gene content (Neobacillus vs. Peribacillus).
The paper isolated representative strains, sequenced their genomes, and showed that the Group 1 strain has a complete denitrification pathway while Group 2 has a partial one.
Evidence: Figure 6a-b shows enzyme presence/absence for the two isolates and across the groups via PICRUSt2.
“Whole genome sequencing of these isolates revealed that, despite their taxonomic proximity, these strains differ in the denitrification enzymes they harbor.”