◆ Microbial Discovery Forge AI Co-Scientist

• Co-Scientist
• Observatory
  • Projects
  • Research Areas
  • Collections
  • Data Explorer
  • Discoveries
  • Community
• Atlas
  • BERIL Atlas
  • Skills
  • Memory: Pitfalls
  • Memory: Performance
  • Research Ideas
• About

Log in with ORCiD

Projects / Metagenome-Prioritized Phage Cocktails for Crohn's Disease and IBD / NB05_tier_a_scoring.ipynb

Nb05 Tier A Scoring

Jupyter notebook from the Metagenome-Prioritized Phage Cocktails for Crohn's Disease and IBD project.

Back to Project

NB05 — Tier-A Scoring Pipeline

Project: ibd_phage_targeting — Pillar 2 close-out Depends on: NB04c (within-substudy meta), NB04e (ecotype-specific Tier-A), NB04h (HMP2 external replication), and the CrohnsPhage mart reference tables (BGC catalog, Kumbhari strain-adaptation, species-IBD associations, phage biology curation).

Purpose

NB04d and NB04e produced the rigor-controlled input candidate set. NB05 applies the Tier-A criteria A3 – A6 from RESEARCH_PLAN.md to produce a scored + ranked target list for Pillar 4 (phage targetability) and Pillar 5 (per-patient cocktails).

Criteria

Criterion	Source	Scoring
A3 Literature + cohort CD-association	NB04c meta + HMP2 replication + ref_cd_vs_hc_differential + ref_species_ibd_associations + ref_phage_biology	0–5
A4 Protective-analog exclusion	NB04c within-substudy sign + curated protective-species list	0 or 1
A5 Engraftment / strain adaptation	Donor 2708 + fact_strain_competition + Kumbhari gene-strain	0 / 0.5 / 1
A6 BGC inflammatory mediator	ref_bgc_catalog + ref_cborf_enrichment	0 / 0.5 / 1

Execution note

This notebook was run via run_nb05.py (standalone script) rather than nbconvert because of an environment-specific numpy.bool serialization issue in the nbconvert notebook-save path. Outputs are authoritative and written to data/nb05_tier_a_scored.tsv, data/nb05_tier_a_verdict.json, and figures/NB05_tier_a_scored.png.

# Imports and constants (see run_nb05.py for full execution)
import warnings; warnings.filterwarnings('ignore')
from pathlib import Path
import json, numpy as np, pandas as pd
import matplotlib.pyplot as plt
DATA_MART = Path.home() / 'data' / 'CrohnsPhage'
DATA_OUT = Path('../data'); FIG_OUT = Path('../figures')

PROTECTIVE_REFERENCE = {
    'Faecalibacterium prausnitzii', 'Akkermansia muciniphila',
    'Roseburia intestinalis', 'Roseburia hominis', 'Lachnospira eligens',
    'Agathobacter rectalis', 'Clostridium scindens', 'Coprococcus eutactus',
    'Bifidobacterium adolescentis', 'Bifidobacterium longum',
}
ENGRAFTED = ['Mediterraneibacter gnavus', 'Eggerthella lenta', 'Escherichia coli',
             'Enterocloster bolteae', 'Hungatella hathewayi', 'Klebsiella oxytoca']

§1. Build candidate set

# See run_nb05.py for full source. Loads NB04e E1 + E3 Tier-A and NB04c engraftment candidates.
# Dedups to unique species with multi-category ecotype_membership string.

E1 Tier-A: 51; E3 provisional Tier-A: 40
Engraftment-confirmed cross-ecotype: 5

Unique candidates: 71
ecotype_membership
E1                                             27
E3_provisional                                 20
E1|E3_provisional                              19
E1|cross_ecotype_engraftment                    4
E1|E3_provisional|cross_ecotype_engraftment     1

§2. Invert synonymy layer (canonical → aliases)

# Required because ref_bgc_catalog, ref_cd_vs_hc_differential etc. use pre-GTDB-r214 names.

  Mediterraneibacter gnavus
    Mediterraneibacter gnavus
    Ruminococcus gnavus
    [Ruminococcus] gnavus
  Enterocloster bolteae
    Clostridium bolteae
    Enterocloster bolteae
  Erysipelatoclostridium innocuum
    Clostridium innocuum
    Erysipelatoclostridium innocuum
    Faecalibacillus faecis
    [Clostridium] innocuum
  Hungatella symbiosa
    Clostridium symbiosum
    Hungatella symbiosa
    Otoolea symbiosa
    [Clostridium] symbiosum

§3. A3 — Literature + cohort CD-association scoring

Five independent signals (each 0 or 1): NB04c confound-free meta, HMP2 external replication concordance, ref_cd_vs_hc_differential (log2FC > 0.5 + FDR < 0.10), ref_species_ibd_associations (dxIBD coefficient > 0 + p < 0.05), ref_phage_biology curated target flag.

# A3 = sum of the 5 signals

A3 score distribution:
a3_score
0    13
1    22
2    25
3     5
4     5
5     1

§4. A4 — Protective-analog exclusion

Binary gate: pass (1) unless the candidate has a negative within-IBD-substudy CD-vs-nonIBD effect (confound-free protective-analog signal) OR sits on the curated protective-species list.

# A4 = 0 (protective-analog risk) or 1 (pass)

A4 score distribution:
a4_score
0     3
1    68

Candidates failing A4:
             species                                                             a4_note
 Anaerostipes hadrus      confound-free effect negative (-0.32) - protective-analog risk
Clostridium scindens on curated protective-species list - strain-level scrutiny required
    Roseburia faecis      confound-free effect negative (-2.74) - protective-analog risk

§5. A5 — Engraftment / strain adaptation

Three evidence tiers: direct donor 2708 engraftment (1.0), Kumbhari strain-competition disease-dominance (0.5), Kumbhari IBD-adapted-strain gene signal (0.5).

# A5 = 0.0 / 0.5 / 1.0

A5 score distribution:
a5_score
0.0    63
0.5     3
1.0     5

A5 = 1.0 direct engraftment:
                  species                          a5_note
        Eggerthella lenta donor 2708 engraftment-confirmed
    Enterocloster bolteae donor 2708 engraftment-confirmed
         Escherichia coli donor 2708 engraftment-confirmed
     Hungatella hathewayi donor 2708 engraftment-confirmed
Mediterraneibacter gnavus donor 2708 engraftment-confirmed

A5 = 0.5 (strain-level signal, top 10):
                     species                                       a5_note
         Anaerostipes hadrus Kumbhari strain-competition disease-dominance
Bacteroides cellulosilyticus Kumbhari strain-competition disease-dominance
      Flavonifractor plautii Kumbhari strain-competition disease-dominance

§6. A6 — BGC inflammatory mediator

For each candidate, look up BGCs in ref_bgc_catalog (via synonymy-inverted matching), count total BGCs and those containing CD-enriched CB-ORFs (from ref_cborf_enrichment, effect > 0.5 + FDR < 0.05). Report MIBiG-matched compound names where available (e.g., E. coli → Colibactin, Yersiniabactin, Enterobactin).

# A6 = 0.0 (no BGC), 0.5 (BGC but no CD-enriched), 1.0 (BGC with CD-enriched CB-ORFs)

BGCs with CD-enriched CB-ORFs: 283

A6 score distribution:
a6_score
0.0    27
0.5    30
1.0    14

A6 = 1.0 (BGC + CD-enriched CB-ORFs, top 15):
                       species                                                                                    a6_note
              Actinomyces oris                                               11 BGCs, 2 CD-enriched CB-ORFs (MIBiG: none)
             Blautia coccoides                                               13 BGCs, 1 CD-enriched CB-ORFs (MIBiG: none)
  Enterocloster asparagiformis                                                4 BGCs, 2 CD-enriched CB-ORFs (MIBiG: none)
       Enterocloster citroniae                                                1 BGCs, 1 CD-enriched CB-ORFs (MIBiG: none)
Enterocloster clostridioformis                                               11 BGCs, 2 CD-enriched CB-ORFs (MIBiG: none)
              Escherichia coli   93 BGCs, 2 CD-enriched CB-ORFs (MIBiG: ['Colibactin', 'Yersiniabactin', 'Enterobactin'])
        Flavonifractor plautii                                                6 BGCs, 1 CD-enriched CB-ORFs (MIBiG: none)
          Hungatella hathewayi                                                6 BGCs, 4 CD-enriched CB-ORFs (MIBiG: none)
     Mediterraneibacter gnavus                                              39 BGCs, 26 CD-enriched CB-ORFs (MIBiG: none)
   Streptococcus parasanguinis                                              51 BGCs, 11 CD-enriched CB-ORFs (MIBiG: none)
      Streptococcus salivarius 98 BGCs, 3 CD-enriched CB-ORFs (MIBiG: ['Salivaricin 9', 'Salivaricin A', 'Cochonodin I'])
           Veillonella atypica                                                2 BGCs, 1 CD-enriched CB-ORFs (MIBiG: none)
            Veillonella dispar                                                1 BGCs, 1 CD-enriched CB-ORFs (MIBiG: none)
           Veillonella parvula                                                8 BGCs, 8 CD-enriched CB-ORFs (MIBiG: none)

§7. Aggregate + rank

# Total Tier-A = A3/5 + A4 + A5 + A6 (0-4 range); actionable threshold = 2.5

Total candidates scored: 71
Actionable (total_score >= 2.5): 6

TOP 20:
 rank                        species                          ecotype_membership  a3_score  a4_score  a5_score  a6_score  total_score  actionable
    1           Hungatella hathewayi                E1|cross_ecotype_engraftment         5         1       1.0       1.0          4.0        True
    2      Mediterraneibacter gnavus E1|E3_provisional|cross_ecotype_engraftment         4         1       1.0       1.0          3.8        True
    3               Escherichia coli                E1|cross_ecotype_engraftment         3         1       1.0       1.0          3.6        True
    4              Eggerthella lenta                E1|cross_ecotype_engraftment         4         1       1.0       0.5          3.3        True
    5         Flavonifractor plautii                           E1|E3_provisional         4         1       0.5       1.0          3.3        True
    6          Enterocloster bolteae                E1|cross_ecotype_engraftment         4         1       1.0       0.0          2.8        True
    7   Enterocloster asparagiformis                           E1|E3_provisional         2         1       0.0       1.0          2.4       False
    8       Streptococcus salivarius                           E1|E3_provisional         2         1       0.0       1.0          2.4       False
    9        Enterocloster citroniae                                          E1         2         1       0.0       1.0          2.4       False
   10 Enterocloster clostridioformis                           E1|E3_provisional         2         1       0.0       1.0          2.4       False
   11              Blautia coccoides                           E1|E3_provisional         2         1       0.0       1.0          2.4       False
   12            Veillonella atypica                           E1|E3_provisional         1         1       0.0       1.0          2.2       False
   13    Streptococcus parasanguinis                           E1|E3_provisional         1         1       0.0       1.0          2.2       False
   14               Actinomyces oris                              E3_provisional         1         1       0.0       1.0          2.2       False
   15            Veillonella parvula                           E1|E3_provisional         1         1       0.0       1.0          2.2       False
   16            Hungatella symbiosa                           E1|E3_provisional         3         1       0.0       0.5          2.1       False
   17               Blautia wexlerae                                          E1         3         1       0.0       0.5          2.1       False
   18   Bacteroides cellulosilyticus                              E3_provisional         0         1       0.5       0.5          2.0       False
   19             Veillonella dispar                                          E1         0         1       0.0       1.0          2.0       False
   20            Streptococcus mitis                                          E1         2         1       0.0       0.5          1.9       False

§8. Scoring matrix heatmap

# Scoring matrix heatmap (top 30) + total-score bar chart.
# Generated by run_nb05.py; saved to figures/NB05_tier_a_scored.png.

§9. Summary

The scored Tier-A list in data/nb05_tier_a_scored.tsv is the final prioritized target set for Pillar 4 (phage targetability) and Pillar 5 (UC Davis per-patient cocktail drafts).

Top 6 actionable candidates (total_score ≥ 2.5, out of 71 scored):

Rank	Species	Ecotype membership	A3	A4	A5	A6	Total	Notes
1	Hungatella hathewayi	E1 | engraftment	5	1	1.0	1.0	4.0	all five A3 signals pass
2	Mediterraneibacter gnavus	E1 | E3_prov | engraftment	4	1	1.0	1.0	3.8	39 BGCs, 26 CD-enriched CB-ORFs
3	Escherichia coli	E1 | engraftment	3	1	1.0	1.0	3.6	MIBiG: Colibactin, Yersiniabactin, Enterobactin
4	Eggerthella lenta	E1 | engraftment	4	1	1.0	0.5	3.3	engrafts + wide IBD literature
5	Flavonifractor plautii	E1 | E3_prov	4	1	0.5	1.0	3.3	Kumbhari strain-competition + BGC
6	Enterocloster bolteae	E1 | engraftment	4	1	1.0	0.0	2.8	engrafts; no BGC hit

Tier-B candidates (score 2.2–2.4, actionable with strain-level or A5 boost): Enterocloster asparagiformis, Streptococcus salivarius (Salivaricin 9/A / Cochonodin I MIBiG matches), E. citroniae, E. clostridioformis, Blautia coccoides, Veillonella atypica, S. parasanguinis, Actinomyces oris, V. parvula.

Candidates failing A4 (protective-analog risk): Anaerostipes hadrus (confound-free effect −0.32), Clostridium scindens (curated protective list), Roseburia faecis (confound-free effect −2.74). These are flagged for strain-level scrutiny before any phage targeting — species-level call alone is insufficient to distinguish pathobiont from protective in these cases.

Caveats (see REPORT.md §5f for full discussion):

• E3 provisional candidates carry single-study (HallAB_2017) evidence within cMD.
• Cross-ecotype engraftment pathobionts are cohort-level; for ecotype-specific dosing, use the ecotype_membership annotation.
• The A3 phage_biology_curated signal catches only 3/71 candidates (curated list is small); most A3 score comes from cross-cohort concordance.

Microbial Discovery Forge

AI co-scientist and research observatory for BERDL-scale microbial discovery.

Maintained by Paramvir S. Dehal

Observatory

• Projects
• Collections
• Discoveries

Atlas

• BERIL Atlas
• Skills
• Pitfalls
• Performance
• Research Ideas

Resources

• BERDL JupyterHub
• Co-Scientist
• How to Cite

Built on BERDL (KBase). Last updated: 2026-06-15 21:31 UTC

×