Carnelian uncovers hidden functional patterns across diverse study populations from whole metagenome sequencing reads

Transcription

1 Carnelian uncovers hidden functional patterns across diverse study populations from whole metagenome sequencing reads Sumaiya Nazeen 1 and Bonnie Berger 1,2* 1 Computer Science and Artificial Intelligence Laboratory (CSAIL), MIT, Cambridge, MA 02139, USA 2 Department of Mathematics, MIT, Cambridge, MA 02139, USA * Corresponding Author: bab@mit.edu Supplementary Information 1

2 Supplementary Table T1. Performance comparison of Carnelian, mi-faser, HUMAnN2-translated, and Kraken2- translated searches on fragments of different lengths with 3% and 5% mutations using mi-faser s gold standard dataset. For shorted reads mi-faser performance is slightly better than Carnelian and HUMAnN2. However, for longer reads, Carnelian achieves better sensitivity and recall. Best performances are marked in red. Mutation Rate Read Length Method Mean Stddev Sensitivity Precision Recall Sensitivity Precision Recall 3% 150-bp Carnelian mi-faser Humann Kraken bp Carnelian mi-faser Humann Kraken bp Carnelian mi-faser Humann Kraken bp Carnelian mi-faser Humann Kraken % 150-bp Carnelian mi-faser Humann Kraken bp Carnelian mi-faser Humann Kraken bp Carnelian mi-faser Humann Kraken bp Carnelian mi-faser Humann Kraken

3 Supplementary Table T2. Performance comparison of Carnelian, mi-faser, HUMAnN2-translated, and Kraken2- translated searches on fragments of different lengths with 3% mutations using Carnelian s gold standard dataset. Carnelian achieves higher sensitivity and F1-score than all other methods with comparable precision. Best performances are marked in read. Read Length Method Mean Stddev Sensitivity Precision Recall Sensitivity Precision Recall 150-bp Carnelian mi-faser Humann Kraken bp Carnelian mi-faser Humann Kraken bp Carnelian mi-faser Humann Kraken

4 Supplementary Table T3. Composition of the synthetic gut metagenome used for functional capacity inference test. Random isolates of the listed species were selected from the ChocoPhlAn database and reads were drawn from their annotated coding sequences ensuring the target coverage in the synthetic metagenome. Species # Reads Proportion Alistipes onderdonkii % Alistipes putredinis % Alistipes shahii % Bacteroides caccae % Bacteroides cellulosilyticus % Bacteroides dorei % Bacteroides massiliensis % Bacteroides ovatus % Bacteroides stercoris % Bacteroides thetaiotaomicron % Bacteroides uniformis % Bacteroides vulgatus % Barnesiella intestinihominis % Dialister invisus % Eubacterium rectale % Faecalibacterium prausnitzii % Parabacteroides distasonis % Parabacteroides merdae % Prevotella copri % Ruminococcus bromii % Total % Uniref_annotated % Uniref_unknown % EC_annotated % 4

5 Supplementary Figure F1. Functional shifts between two sets predicted by Carnelian, mi-faser, HUMAnN2- translated, and Kraken2-translated searches in the functional inference dataset from Lindgreen et al. study [1]. A positive log fold change means an increase in set A relative to set B and vice versa. The expected fold change amounts were given by the original paper based on the taxonomic differences of the two sets. The test datasets were created with differences in the relative abundance of cyanobacteria (photosynthesis; more abundant in set A), Bradyrhizobium and Rhizobium (nitrogen fixation; more abundant in set A), and known pathogens (more abundant in set B). We profiled the metagenomes from the two sets with all four methods and calculated pathway abundances by mapping the identified ECs to the carbon-fixation, photosynthesis, nitrogen-fixation, two-component systems, bacterial chemotaxis, and cell motility pathways and summing the abundances. Carnelian identifies the expected direction of change in each of the three categories, where rest of the methods don t. 5

6 Supplementary Table T4. Composition of the dataset used for testing functional change inference. Two sets of six complex metagenomes were created with varying proportions of coding sequences from 20 different species of proteobacteria, cyanobacteria, photosynthetic bacteria, nitrogen-fixing bacteria and known pathogens. Species Set A Set B Average Proportions A1 A2 A3 B1 B2 B3 Set A Set B Anabaena sp % 15% Bradyrhizobium % 1% diazoefficiens Bradyrhizobium elkanii % 1% Campylobacter coli % 2% Chlorobium chlorochromatii % 2% Chloroflexus aurantiacus % 2% Erythrobacter litoralis % 2% Escherichia albertii % 5% Escherichia coli % 5% Helicobacter canadensis % 5% Microcystis aeruginosa % 13% Nodularia spumigena % 15% Nostoc punctiforme % 17% Rhizobium freirei % 1% Rhizobium grahamii % 1% Rhodomicrobium vannielii % 1% Rhodospirillum centenum % 2% Salmonella enterica % 4% Trichodesmium erythraeum % 1% Vibrio campbellii % 5% Total % 100% Uniref annotated % 85% Uniref unknown % 15% EC annotated % 9% 6

7 Supplementary Table T5. Inference of functional changes from the two-set complex metagenomes. Carnelian achieves slightly higher sensitivity at both EC and pathway level compared to mi-faser and significantly higher sensitivity than HUMAnN2-translated and Kraken2-translated searches. Sensitivity is measured as the proportion of functional terms identified by a method with correct direction of shift between two sets. Method Sensitivity at EC level (%) Sensitivity at pathway level (%) All ECs Highly variable ECs (expected abs(logfc) > 1) All pathways Highly variable Pathways (expected abs(logfc) > 0.58) Carnelian mi-faser HUMAnN2-translated Kraken2-translated

8 Supplementary Table T6. Significantly differentially abundant ECs identified by Carnelian in the T2D-Qin dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-mean ctrl-mean logfc adj pval

9 Supplementary Table T7. Significantly differentially abundant ECs identified by mi-faser in the T2D-Qin dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC fold-change logfc adj pval EC fold-change logfc adj pval

10 Supplementary Table T8. Significantly differentially abundant ECs identified by HUMAnN2-translated search in the T2D-Qin dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC fold-change logfc adj pval

11 Supplementary Table T9. Significantly differentially abundant ECs identified by Kraken2-translated search in the T2D-Qin dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC fold-change logfc adj pval EC foldchange logfc adj pval n

12 Supplementary Table T10. Significantly differentially abundant ECs identified by Carnelian in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ECs marked with * are significantly variable between IGT and NGT individuals as well. EC t2d-mean igt-mean ngt-mean t2d-ngt logfc t2d-ngt adj pval igt-ngt logfc igt-ngt adj pval * * n * * * * *

13 Supplementary Table T11. Significantly differentially abundant ECs identified by mi-faser in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt-fc t2d-ngtlogfc t2d-ngt-adj-p igt-ngt-fc igt-ngt-logfc igt-ngt-adj-p

14 Supplementary Table T11. Significantly differentially abundant ECs identified by mi-faser in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt-fc t2d-ngtlogfc t2d-ngt-adjlogfc igt-ngt-fc igt-ngt igt-ngt-adjp

15 Supplementary Table T11 (continued). Significantly differentially abundant ECs identified by mi-faser in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt-fc t2d-ngtlogfc t2d-ngt-adj-p igt-ngt-fc igt-ngt-logfc igt-ngt-adj-p

16 Supplementary Table T11 (continued). Significantly differentially abundant ECs identified by mi-faser in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt-fc t2d-ngtlogfc t2d-ngt-adj-p igt-ngt-fc igt-ngt-logfc igt-ngt-adj-p

17 Supplementary Table T12. Significantly differentially abundant ECs identified by HUMAnN2-translated search in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt t2d-ngt t2d-ngt igt-ngt igt-ngt igt-ngt FC logfc adj-p FC logfc adj-p n

18 Supplementary Table T12 (continued). Significantly differentially abundant ECs identified by HUMAnN2- translated search in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt t2d-ngt t2d-ngt igt-ngt igt-ng igt-ngt FC logfc adj-p FC logfc adj-p

19 Supplementary Table T13. Significantly differentially abundant ECs identified by Kraken2-translated search in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt-fc t2d-ngtlogfc t2d-ngt-adjlogfc igt-ngt-fc igt-ngt igt-ngt-adjp

20 Supplementary Table T13 (continued). Significantly differentially abundant ECs identified by Kraken2-translated search in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt FC t2d-ngt logfc t2d-ngt adj-p igt-ngt FC igt-ngt logfc igt-ngt adj-p

21 Supplementary Table T13 (continued). Significantly differentially abundant ECs identified by Kraken2-translated search in the T2D-Karlsson dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC t2d-ngt-fc t2d-ngtlogfc t2d-ngt-adjlogfp igt-ngt-fc igt-ngt- igt-ngt-adj

22 Supplementary Table T14. Significantly differentially abundant ECs identified by Carnelian in the PD-Bedarf dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC pd_mean ctrl_mean fold-change logfc adj pval

23 Supplementary Table T14 (continued). Significantly differentially abundant ECs identified by Carnelian in the PD-Bedarf dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC pd_mean ctrl_mean fold-change logfc adj pval

24 Supplementary Table T15. Significantly differentially abundant ECs identified by mi-faser in the PD-Bedarf dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC fold-change logfc adj pval EC fold-change logfc adj pval n

25 Supplementary Table T15 (continued). Significantly differentially abundant ECs identified by mi-faser in the PD- Bedarf dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC fold-change logfc adj p EC fold-change logfc adj p

26 Supplementary Table T16. Significantly differentially abundant ECs identified by HUMAnN2-translated search in the PD-Bedarf dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC fold-change logfc adj p EC fold-change logfc adj p n n

27 Supplementary Table T16 (continued). Significantly differentially abundant ECs identified by HUMAnN2- translated search in the PD-Bedarf dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p- value < 0.05 and abs (log fold change) > EC fold-change logfc adj pval EC fold-change logfc adj pval

28 Supplementary Table T17. Significantly differentially abundant ECs identified by Kraken2-translated search in the PD-Bedarf dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC fold-change logfc adj pval EC fold-change logfc adj pval

29 Supplementary Table T17 (continued). Significantly differentially abundant ECs identified by Kraken2-translated search in the PD-Bedarf dataset. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > EC fold-change logfc adj pval

30 (a) T2D-Qin dataset (b) T2D-Karlsson dataset (c) PD-Bedarf dataset Supplementary Figure F2. The overlap between significantly variable ECs identified by Carnelian, mi-faser, HUMAnN2-translated and Kraken2-translated searches. (a) T2D-Qin dataset, (b) T2D-Karlsson dataset, and (c) PD-Bedarf dataset. 30

31 Supplementary Figure F3. Classification of patients vs controls in Chinese and European T2D cohorts using the combined set of markers identified by Carnelian, mi-faser, HUMAnN2-translated, and Kraken2- translated. Using Carnelian-identified highly variable ECs, we can consistently achieve ~80% area under the curve (AUC) on average on both T2D-Qin and T2D-Karlsson datasets, whereas using the highly variable ECs identified by other methods, an average AUC of 73-76% can be achieved on both datasets. 31

32 Supplementary Table T18. Combined EC markers identified by Carnelian that can classify T2D patients vs controls in both Chinese and European population with ~80% area under the curve on average n

33 Supplementary Table T19. Pathways identified as significantly variable between T2D patients and healthy controls in the T2D-Qin dataset using Carnelian-generated functional profiles. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name fold-change logfc adj pval Glycolysis / Gluconeogenesis Citrate cycle (TCA cycle) Pentose phosphate pathway Pentose and glucuronate interconversions Fructose and mannose metabolism Galactose metabolism Ascorbate and aldarate metabolism Fatty acid biosynthesis Fatty acid elongation Steroid biosynthesis Oxidative phosphorylation Photosynthesis Pyrimidine metabolism Alanine, aspartate and glutamate metabolism Aflatoxin biosynthesis Valine, leucine and isoleucine biosynthesis Arginine and proline metabolism Tyrosine metabolism Phenylalanine metabolism Glutathione metabolism Amino sugar and nucleotide sugar metabolism Arachidonic acid metabolism Sphingolipid metabolism Glycosphingolipid biosynthesis - globo and isoglobo series Pyruvate metabolism Thiamine metabolism Vitamin B6 metabolism Nicotinate and nicotinamide metabolism Biotin metabolism Drug metabolism - other enzymes

34 Supplementary Table T20. Pathways identified as significantly variable between T2D patients and NGT individuals in the T2D-Karlsson dataset using Carnelian-generated functional profiles. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name t2d-ngt t2d-ngt t2d-ngt igt-ngt igt-ngt igt-ngt FC logfc adj pval FC logfc adj-pval Pentose phosphate pathway Pentose and glucuronate interconversions Fructose and mannose metabolism Galactose metabolism Fatty acid biosynthesis Oxidative phosphorylation Photosynthesis Arginine biosynthesis Alanine, aspartate and glutamate metabolism Glycine, serine and threonine metabolism Valine, leucine and isoleucine biosynthesis Lysine biosynthesis Carbapenem biosynthesis Histidine metabolism Phenylalanine, tyrosine and tryptophan biosynthesis Phenazine biosynthesis Selenocompound metabolism Cyanoamino acid metabolism D-Arginine and D-ornithine metabolism Other glycan degradation Glycerolipid metabolism Arachidonic acid metabolism Sphingolipid metabolism Glycosphingolipid biosynthesis ganglio series Thiamine metabolism Aminoacyl-tRNA biosynthesis Drug metabolism - other enzymes

35 Supplementary Table T20 (continued). Pathways identified as significantly variable between T2D patients and NGT individuals in the T2D-Karlsson dataset using Carnelian-generated functional profiles. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name t2d-ngt t2d-ngt t2d-ngt igt-ngt igt-ngt igt-ngt FC logfc adj pval FC logfc adj-pval Atrazine degradation Retinol metabolism C5-Branched dibasic acid metabolism Terpenoid backbone biosynthesis Nitrogen metabolism Sulfur metabolism Polycyclic aromatic hydrocarbon degradation Chloroalkane and chloroalkene degradation Naphthalene degradation

36 Supplementary Table T21. Pathways identified as significantly variable between T2D patients and healthy controls in the T2D-Qin dataset using mi-faser-generated functional profiles. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name foldchange logfc adj-pval Fatty acid biosynthesis Photosynthesis Aflatoxin biosynthesis D-Arginine and D-ornithine metabolism Glutathione metabolism Various types of N-glycan biosynthesis Biosynthesis of 12-, 14- and 16-membered macrolides Lipoarabinomannan (LAM) biosynthesis New! Glycosphingolipid biosynthesis - globo and isoglobo series Aminobenzoate degradation Riboflavin metabolism Vitamin B6 metabolism Lipoic acid metabolism Carotenoid biosynthesis Flavone and flavonol biosynthesis Insect hormone biosynthesis Biosynthesis of secondary metabolites - unclassified New! Biosynthesis of ansamycins

37 Supplementary Table T22. Pathways identified as significantly variable between T2D patients and NGT individuals in the T2D-Karlsson dataset using mi-faser-generated functional profiles. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name t2d-ngt t2d-ngt t2d-ngt igt-ngt igt-ngt igt-ngt FC logfc adj pval FC logfc adj-pval Oxidative phosphorylation Photosynthesis Arginine biosynthesis Monobactam biosynthesis Valine, leucine and isoleucine biosynthesis Phenylalanine, tyrosine and tryptophan biosynthesis Cyanoamino acid metabolism Other glycan degradation Glycosaminoglycan degradation Glycerolipid metabolism Arabinogalactan biosynthesis Mycobacterium New! Sphingolipid metabolism Glycosphingolipid biosynthesis ganglio series Atrazine degradation

38 Supplementary Table T23. Pathways identified as significantly variable between T2D patients and healthy controls in the T2D-Qin dataset using functional profiles generated by HUMAnN2-translated search. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name foldchange logfc adj pval Citrate cycle (TCA cycle) Synthesis and degradation of ketone bodies Oxidative phosphorylation Photosynthesis Tryptophan metabolism Glycerolipid metabolism Glycosphingolipid biosynthesis - globo and isoglobo series Toluene degradation Nitrotoluene degradation Vitamin B6 metabolism Terpenoid backbone biosynthesis Biosynthesis of secondary metabolites unclassified Supplementary Table T24. Pathways identified as significantly variable between T2D patients and NGT individuals in the T2D-Karlsson dataset using functional profiles generated by HUMAnN2-translated search. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name t2d-ngt t2d-ngt t2d-ngt igt-ngt igt-ngt igt-ngt FC logfc adj pval FC logfc adj-pval Fructose and mannose metabolism Fatty acid degradation Lysine degradation Histidine metabolism Cyanoamino acid metabolism One carbon pool by folate Atrazine degradation

39 Supplementary Table T25. Pathways identified as significantly variable between T2D patients and healthy controls in the T2D-Qin dataset using functional profiles generated by Kraken2-translated search. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name foldchange logfc adj pval Fatty acid biosynthesis Photosynthesis Aflatoxin biosynthesis Tryptophan metabolism Novobiocin biosynthesis Glutathione metabolism Glycosaminoglycan degradation Sphingolipid metabolism Glycosphingolipid biosynthesis - globo and isoglobo series Nicotinate and nicotinamide metabolism Biosynthesis of unsaturated fatty acids Biosynthesis of ansamycins Supplementary Table T26. Pathways identified as significantly variable between T2D patients and NGT individuals in the T2D-Karlsson dataset using functional profiles generated by HUMAnN2-translated search. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name t2d-ngt t2d-ngt t2d-ngt igt-ngt igt-ngt igt-ngt FC logfc adj pval FC logfc adj-pval Ascorbate and aldarate metabolism Synthesis and degradation of ketone bodies Oxidative phosphorylation Photosynthesis Arginine biosynthesis Aflatoxin biosynthesis Monobactam biosynthesis Fluorobenzoate degradation Selenocompound metabolism Mannose type O-glycan biosynthesis Glycerolipid metabolism Chloroalkane and chloroalkene degradation C5-Branched dibasic acid metabolism Atrazine degradation

40 Supplementary Table T27. Pathways identified as significantly variable between PD patients and healthy controls in the PD-Bedarf dataset using Carnelian-generated functional profiles. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name Fold logfc adjusted Change p-value Pentose and glucuronate interconversions Fructose and mannose metabolism Galactose metabolism Ascorbate and aldarate metabolism Alanine, aspartate and glutamate metabolism Benzoate degradation Phenylalanine, tyrosine and tryptophan biosynthesis D-Alanine metabolism Glutathione metabolism Starch and sucrose metabolism Streptomycin biosynthesis Polyketide sugar unit biosynthesis Glycosaminoglycan degradation Peptidoglycan biosynthesis Lipoarabinomannan (LAM) biosynthesis Sphingolipid metabolism Glycosphingolipid biosynthesis - globo and isoglobo series Glycosphingolipid biosynthesis - ganglio series Folate biosynthesis Porphyrin and chlorophyll metabolism Zeatin biosynthesis Drug metabolism - other enzymes Biosynthesis of secondary metabolites - unclassified Biosynthesis of vancomycin group antibiotics

41 Supplementary Table T28. Pathways identified as significantly variable between PD patients and healthy controls in the PD-Bedarf dataset using mi-faser-generated functional profiles. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name Fold Change logfc adj pval Pentose and glucuronate interconversions Fructose and mannose metabolism Galactose metabolism Ascorbate and aldarate metabolism Fatty acid elongation Ubiquinone and other terpenoid-quinone biosynthesis Purine metabolism Pyrimidine metabolism Phenazine biosynthesis Phosphonate and phosphinate metabolism Selenocompound metabolism D-Arginine and D-ornithine metabolism Glutathione metabolism Starch and sucrose metabolism Other glycan degradation Glycosaminoglycan degradation Inositol phosphate metabolism Sphingolipid metabolism Glycosphingolipid biosynthesis - ganglio series Nitrotoluene degradation One carbon pool by folate Biotin metabolism Folate biosynthesis Limonene and pinene degradation Zeatin biosynthesis Caprolactam degradation Aminoacyl-tRNA biosynthesis Drug metabolism - other enzymes Steroid degradation Biosynthesis of secondary metabolites unclassified

42 Supplementary Table T29. Pathways identified as significantly variable between PD patients and healthy controls in the PD-Bedarf dataset functional profiles generated by HUMAnN2-translated search. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name Fold logfc Adjusted Change Wilcox p-value Citrate cycle (TCA cycle) Pentose phosphate pathway Pentose and glucuronate interconversions Galactose metabolism Synthesis and degradation of ketone bodies Purine metabolism Pyrimidine metabolism Monobactam biosynthesis Valine, leucine and isoleucine degradation Lysine degradation beta-alanine metabolism D-Arginine and D-ornithine metabolism Starch and sucrose metabolism Pyruvate metabolism Nitrotoluene degradation Propanoate metabolism Butanoate metabolism One carbon pool by folate Methane metabolism Carbon fixation in photosynthetic organisms Carbon fixation pathways in prokaryotes Pantothenate and CoA biosynthesis

43 Supplementary Table T30. Pathways identified as significantly variable between PD patients and healthy controls in the PD-Bedarf dataset functional profiles generated by Krake2-translated search. Significance thresholds used: BH corrected Wilcoxon ranksum test p-value < 0.05 and abs (log fold change) > ID Name Fold Change logfc Adjusted Wilcox p-value Pentose and glucuronate interconversions Fructose and mannose metabolism Galactose metabolism Ascorbate and aldarate metabolism Primary bile acid biosynthesis Steroid hormone biosynthesis Purine metabolism Pyrimidine metabolism Alanine, aspartate and glutamate metabolism Arginine and proline metabolism Histidine metabolism Phenylalanine, tyrosine and tryptophan biosynthesis Selenocompound metabolism D-Arginine and D-ornithine metabolism Glutathione metabolism Starch and sucrose metabolism Other glycan degradation Amino sugar and nucleotide sugar metabolism Glycosaminoglycan degradation Lipopolysaccharide biosynthesis Peptidoglycan biosynthesis Glycosphingolipid biosynthesis - ganglio series Toluene degradation Riboflavin metabolism Folate biosynthesis Zeatin biosynthesis Nitrogen metabolism Aminoacyl-tRNA biosynthesis Biosynthesis of secondary metabolites - unclassified

44 (a) EC profiles (b) Pathway profiles Supplementary Figure F4. Agreement between the functional profiles generated by Carnelian from the Roche 454 and Illumina GA data from the same DNA sample. (a) Agreement between EC profiles; (b) Agreement between pathway profiles. 44

45 Supplementary Table T31. Kendall rank correlation between the functional profiles of VAG-pond samples annotated by Carnelian. Epilimnion (Surface) Metalimnion (Middle) Hypolimnion (Bottom) Freshwater (Control) Epilimnion Metalimnion Hypolimnion Freshwater 1-S 2-S 3-S 1-M 2-M 3-M 2-B 3-B 1-F 2-F 1-S S S M M M B B F F

46 Supplementary Table T32. Highly variable ECs between the epilimnion, metalimnion, and hypolimnion layers found by Carnelian in the VAG-pond dataset. EC EpMe_FC EpHy_FC MeHy_FC EC EpMe_FC EpHy_FC Me_Hy_FC n

47 Supplementary Table T33. Highly variable pathways between the epilimnion, metalimnion, and hypolimnion layers found by Carnelian in the VAG-pond dataset. Thresholds used: coverage >= 0.30 and absolute log foldchange > Here, coverage is calculated as the ratio of the number of Carnelian-identified ECs mapped to a pathway to the total number of gold standard ECs in the pathway. ID Pathway coverage lfcsm lfcsb lfcmb Synthesis and degradation of ketone bodies Monobactam biosynthesis Geraniol degradation Benzoate degradation Lipoarabinomannan (LAM) biosynthesis New! Carbapenem biosynthesis Biosynthesis of unsaturated fatty acids Arabinogalactan biosynthesis - Mycobacterium New! Valine, leucine and isoleucine degradation Valine, leucine and isoleucine biosynthesis alpha-linolenic acid metabolism Lysine degradation C5-Branched dibasic acid metabolism Butanoate metabolism Phosphonate and phosphinate metabolism Terpenoid backbone biosynthesis Fatty acid degradation Lysine biosynthesis Dioxin degradation Ubiquinone and other terpenoid-quinone biosynthesis Ethylbenzene degradation Steroid degradation Glycine, serine and threonine metabolism Lipoic acid metabolism Galactose metabolism Penicillin and cephalosporin biosynthesis Oxidative phosphorylation Starch and sucrose metabolism Other glycan degradation Mannose type O-glycan biosynthesis D-Arginine and D-ornithine metabolism

48 Supplementary Figure F5. Non-metric multidimensional scaling (NMDS) plot depicting the Carnelian-derived functional profiles of beach sand metagenomes from the DWH-spill dataset. Samples in each phase are functionally similar to each other. Samples in the oil-phase are functionally more similar to the samples in the post-oil phase. 48

49 Supplementary Table T34. Hydrocarbon-degrading ECs involved in BTEX metabolism pathways found enriched in the oil phase compared to the pre-oil phase in DWH-spill dataset by Carnelian. EC Number Name Oil_by_Pre Oil_by_Post hydroxyacyl-CoA dehydrogenase Sulfhydrogenase 1 subunit beta Catechol 1,2-dioxygenase Catechol-2,3-dioxygenase Protocatechuate 3,4-dioxygenase beta chain Manganese-dependent 2,3-dihydroxybiphenyl 1,2-dioxygenase Gallate dioxygenase * Nitronate monooxygenase Alpha-ketoglutarate-dependent taurine dioxygenase * Anthranilate 1,2-dioxygenase large subunit p-hydroxybenzoate hydroxylase (PHBH) hydroxybenzoate 6-hydroxylase Phenol 2-monooxygenase Cytochrome P450 3A Alkanesulfonate monooxygenase Cytochrome p450 CYP199A Ammonia monooxygenase alpha subunit (AMO) Acetaldehyde dehydrogenase Phenylacetaldehyde dehydrogenase (PAD) Maleylacetate reductase * Cyclohex-1-ene-1-carbonyl-CoA dehydrogenase (Ch1CoA) * Cyclohexane-1-carbonyl-CoA dehydrogenase (ChCoA) Pyrogallol hydroxytransferase large subunit ketoacyl-CoA thiolase Acetyl-CoA acetyltransferase A Glutaconate CoA-transferase subunit A oxoadipate CoA-transferase subunit A Acetate CoA-transferase subunit alpha oxoadipate enol-lactonase Putative carboxymethylenebutenolidase Aryldialkylphosphatase Acetamidase Nitrilase (S)-2-haloacid dehalogenase Haloacetate dehalogenase H Haloalkane dehalogenase chlorobenzoyl coenzyme A dehalogenase hydroxybenzoate decarboxylase subunit C Benzoylformate decarboxylase (BFD) Glutaconyl-CoA decarboxylase subunit gamma carboxy-4-hydroxy-2-oxoadipic acid aldolase hydroxy-2-oxovalerate aldolase (HOA) enoyl-coa hydratase keto-4-pentenoate hydratase oxalmesaconate hydratase (OMA hydratase)

50 Supplementary Table T34 (continued). Hydrocarbon-degrading ECs involved in BTEX metabolism pathways found enriched in the oil phase compared to the pre-oil phase in the DWH-spill dataset by Carnelian. EC Number Name Oil_by_Pre Oil_by_Post Mandelate racemase (MR) oxalomesaconate tautomerase Muconolactone Delta-isomerase (MIase) hydroxylaminophenol mutase (3HAP mutase) carboxy-cis,cis-muconate cycloisomerase Chloromuconate cycloisomerase Anthranilate--CoA ligase Supplementary Table T35. Highly variable hydrocarbon metabolism pathways found by Carnelian in the DWHspill dataset. Here, coverage is calculated as the ratio of the number of Carnelian-identified ECs mapped to a pathway to the total number of gold standard ECs in the pathway. Pathways having coverage > 0.30 are reported. ID Pathway Oil-Pre Oil-Post # ECs Coverage Fold Change Fold Change mapped Benzoate degradation Aminobenzoate degradation Chloroalkane and chloroalkene degradation Chlorocyclohexane and chlorobenzene degradation Dioxin degradation Toluene degradation Fluorobenzoate degradation Styrene degradation Xylene degradation Ethylbenzene degradation Polycyclic aromatic hydrocarbon degradation

51 Supplementary Table T36. Run-time and memory requirement of Carnelian vs mi-faser, HUMAnN2-translated, and Kraken2-translated searches. The reads were generated from the synthetic gut metagenome we created using the 20 most abundant species from HMP project. The datasets had only a small fraction of reads with known EC labels. Although Carnelian predicts possible ORFs from >80% of the reads, with the current small EC database, it can annotate only a fraction of the reads. Still the number of reads annotated by Carnelian are higher than other methods. 150 bp paired-end 10M (~820k with known EC labels) Carnelian mi-faser Read length # Reads Method User CPU time MRSS (GB) # annotated (min) 250 bp 5M Carnelian single-end (~450k with mi-faser known EC Humann labels) translated Kraken2- translated Humann2- translated Kraken2- translated

52 Supplementary Table T37. Performance of Carnelian with reduced-size amino acid alphabets on our crossvalidation test set containing ~3M 100 bp fragments. Model Alphabet Size Sensitivity (%) Precision (%) F1-score (%) Peak Memory (GB) (AA) Full Alphabet MWL MWL MWL Physico-Chemical HP Model mi-faser, HUMAnN MWL2000: Murphy, L. R., Wallqvist, A., & Levy, R. M. (2000). Simplified amino acid alphabets for protein fold recognition and implications for folding. Protein Engineering, 13(3), Physio-Chemical: Amino acids grouped according to 5 physico-chemical properties A (Aliphatic): IVL, R (aromatic): FYWH, C (Charged): KRDE, T (Tiny): GACS, D (Diverse): TMQNP 3 HP Model: Groups amino acids as polar (hydrophilic) or hydrophobic P: AGTSNQDEHRKP, H: CMFILVWY 4 Diamond aligner, used by mi-faser and HUMAnN2-translated, inherently represents the proteins in its database with a reduced amino acid alphabet of size

53 Supplementary Figure F6. Example of low-density even coverage hashing representation of an amino acid k- mer. A (k, r)-hash function can be thought of as a binary vector of length k with r 1's where each 1 indicates a marked position in the k-mer. Leveraging the hashing technique from Opal, Carnelian starts with a set of hash functions, represented as a hash matrix where the first row has 1's in first r positions, the second row has 1's in second r positions, and so on. Carnelian then permutes the columns of this matrix repeatedly to generate even coverage LSH functions. The rows then give the corresponding hashes of a k-mer. 53