Mouse Phenome Database

jaechang-hits/mouse-phenome-database

Retrieve mouse phenotype data from the Jackson Laboratory Mouse Phenome Database (MPD) via its REST API. Browse 520+ projects, look up per-project measure metadata, pull strain-level means (raw or LS-mean adjusted) and per-animal values, find measures by MP/VT ontology terms, and resolve strain nomenclature or gene coordinates. Use for QTL support, cross-strain comparison, mouse model selection, and ontology-driven phenotype discovery. Use monarch-database for disease-gene-phenotype knowledge graphs; ensembl-database for mouse genome annotations.

This is a copy. The original lives at biotender-max/mouse-phenome-database.

7k tokens

context cost

the whole folder, loaded on every use

files

instructions only

copies elsewhere

how many repositories repackaged it

294

stars on the repo

on the repository, not the skill itself

Install

one command, takes just this skill from the repository

npx skills add https://github.com/jaechang-hits/SciAgent-Skills --skill mouse-phenome-database

The instruction itself

33 sections, as written by the author

mouse-phenome-database

Overview

The Mouse Phenome Database (MPD), maintained at the Jackson Laboratory, catalogs standardized phenotype measurements across inbred, recombinant inbred (e.g., BXD), and Collaborative Cross / Diversity Outbred mouse panels. It aggregates 520+ projects spanning metabolic, cardiovascular, behavioral, hematological, and immunological traits. The REST API at https://phenome.jax.org/api is free, requires no authentication, and is documented at <https://phenome.jax.org/about/api>. MPD measurement IDs (measnum) are project-scoped 5-digit integers — there is no global "measnum 10001 = body weight" mapping; valid measnums must be discovered per project via the measureinfo endpoint.

When to Use

Selecting inbred strains with extreme phenotypes (highest/lowest fasted glucose, body weight, heart rate, etc.) as experimental models
Pulling individual-animal data from BXD / CC / DO panels for QTL mapping with R/qtl2 or similar tools
Comparing strain means and variance across metabolic, behavioral, or cardiovascular measures for genetic background studies
Finding MPD projects that measure a trait of interest using ontology terms (MP, VT, MA) or free-text descriptions
Validating mouse strain nomenclature (canonical JAX names ↔ stock numbers ↔ MGI IDs) before submitting orders or analyses
Looking up coordinates and annotations for mouse genes in the MPD/MGI cross-reference
Use monarch-database instead for disease-gene-phenotype knowledge graphs (HPO ↔ MP ↔ disease)
Use ensembl-database instead for transcript-level mouse gene annotations and variant consequence prediction

Prerequisites

Python packages: requests, pandas, matplotlib
Data requirements: a project symbol (e.g., Jaxwest1, Auwerx1) or a measnum (e.g., 15101); strain names follow JAX canonical nomenclature (e.g., C57BL/6J, DBA/2J)
Environment: internet connection; no API key required
Rate limits: no published hard limit; keep bursts under ~5 requests/second and add time.sleep(0.3) between requests in loops

pip install requests pandas matplotlib

Quick Start

import requests

MPD = "https://phenome.jax.org/api"

# 1) Pick a project (Jaxwest1 — cardiovascular phenotyping on inbred panel)
r = requests.get(f"{MPD}/projects/Jaxwest1/strains", timeout=30)
strains = r.json()["strains"]
print(f"Jaxwest1: {len(strains)} strains tested")

# 2) Discover its measures
r = requests.get(f"{MPD}/pheno/measureinfo/Jaxwest1", timeout=30)
measures = r.json()["measures_info"]
print(f"Jaxwest1 measures: {len(measures)}; first: measnum={measures[0]['measnum']} "
      f"varname={measures[0]['varname']}  ({measures[0]['descrip']}, {measures[0]['units']})")

# 3) Pull strain means for heart rate (varname=HR, measnum=15101)
r = requests.get(f"{MPD}/pheno/strainmeans/15101", timeout=30)
sm = r.json()["strainmeans"]
print(f"\nHeart rate strain means: {len(sm)} rows  (one per strain × sex)")
top = sorted(sm, key=lambda x: x["mean"], reverse=True)[:5]
for s in top:
    print(f"  {s['strain']:<20}  sex={s['sex']}  mean={s['mean']:.0f} {s.get('varname','')}  n={s['nmice']}")

Core API

Module 1: Browse Projects — `/projects`

Lists all MPD projects with full metadata. Filter via investigator, projsym, projid, mpdsector, largecollab, panelsym. Use /project_filters/{filtername} to see the allowed values of mpdsector, largecollab, or panelsym before filtering.

import requests, pandas as pd

MPD = "https://phenome.jax.org/api"

# List allowed panel symbols (e.g., BXD, CC, DO)
filters = requests.get(f"{MPD}/project_filters/panelsym", timeout=30).json()
print(f"Available panels ({filters['count']}):", [t['term'] for t in filters['terms']][:10])

# All projects in the BXD recombinant inbred panel
r = requests.get(f"{MPD}/projects", params={"panelsym": "BXD"}, timeout=30)
projects = r.json()["projects"]
print(f"BXD projects: {len(projects)}")
df = pd.DataFrame([{
    "projsym": p["projsym"],
    "pi": p.get("pistring", "")[:40],
    "nstrains": p.get("nstrains"),
    "ages": p.get("ages"),
    "sector": p.get("mpdsector"),
    "title": (p.get("title") or "")[:60],
} for p in projects])
print(df.head(10).to_string(index=False))

# Filter by MPD sector — komp, pheno, qtla, snp, onestrain, phenoarchive
r = requests.get(f"{MPD}/projects", params={"mpdsector": "qtla"}, timeout=30)
qtl_projects = r.json()["projects"]
print(f"QTL-archive projects: {len(qtl_projects)}")
for p in qtl_projects[:5]:
    print(f"  {p['projsym']:<15} panel={p.get('panelsym') or '--':<6} nstrains={str(p.get('nstrains') or '--'):>4}  {(p.get('title') or '')[:55]}")

Module 2: Project Detail — `/projects/{projsym}/...`

Each project has sub-resources for its dataset (CSV of every animal × every measure), the strain panel it tested, the publications it produced, and (for QTL projects) the genetic markers used.

import requests, io, pandas as pd

MPD = "https://phenome.jax.org/api"

# Full per-animal dataset as CSV (default). Use json=yes for JSON.
r = requests.get(f"{MPD}/projects/Jaxwest1/dataset", timeout=60)
df = pd.read_csv(io.StringIO(r.text))
print(f"Jaxwest1 dataset: {df.shape[0]} animals × {df.shape[1]} columns")
print(df.columns[:12].tolist())
print(df[["strain", "sex", "animal_id", "HR", "QRS", "bw"]].head(5).to_string(index=False))

# Strains tested in a project + publication list
strains = requests.get(f"{MPD}/projects/Jaxwest1/strains", timeout=30).json()
print(f"Jaxwest1 strains ({strains['count']}):")
for s in strains["strains"][:5]:
    print(f"  {s['strainname']:<20}  stock={s['stocknum']}  vendor={s['vendor']}")

pubs = requests.get(f"{MPD}/projects/Jaxwest1/publications", timeout=30).json()
print(f"\nPublications: {pubs['count']}")

Module 3: Measure Discovery — `/pheno/measureinfo/{selector}`

This is the canonical way to discover valid measnum values. The selector is either a project symbol (returns all measures in that project) or a measnum (returns metadata for one measure).

import requests, pandas as pd

MPD = "https://phenome.jax.org/api"

# All measures in the Jaxwest1 cardiovascular project
r = requests.get(f"{MPD}/pheno/measureinfo/Jaxwest1", timeout=30)
measures = r.json()["measures_info"]
df = pd.DataFrame([{
    "measnum": m["measnum"],
    "varname": m["varname"],
    "descrip": m["descrip"],
    "units": m.get("units"),
    "sex": m.get("sextested"),
    "age": m.get("ageweeks"),
} for m in measures])
print(f"Jaxwest1 has {len(df)} measures")
print(df.head(10).to_string(index=False))

# Single-measure metadata lookup (protocol + dimensional details)
r = requests.get(f"{MPD}/pheno/measureinfo/15101", timeout=30).json()
m = r["measures_info"][0]
print(f"measnum {m['measnum']} ({m['varname']}): {m['descrip']}")
print(f"  units: {m.get('units')}")
print(f"  project: {m.get('projsym')}  panel: {m.get('panelsym') or m.get('paneldesc')}")
print(f"  sex tested: {m.get('sextested')}  age: {m.get('ageweeks')}")
print(f"  method:  {(m.get('method') or '')[:120]}")

Module 4: Strain Means — `/pheno/strainmeans/{selector}`

Returns strain × sex summary statistics. The selector takes a project symbol (all strain means for the project) or one-or-more comma-separated measnums. Each row contains measnum, varname, strain, strainid, sex, mean, sd, sem, cv, minval, maxval, nmice, zscore.

import requests, pandas as pd

MPD = "https://phenome.jax.org/api"

# Strain means for one measure (heart rate, measnum=15101 from Jaxwest1)
r = requests.get(f"{MPD}/pheno/strainmeans/15101", timeout=30)
sm = pd.DataFrame(r.json()["strainmeans"])
print(f"Rows: {len(sm)}  ({sm['strain'].nunique()} strains × {sm['sex'].nunique()} sexes)")

# Rank strains by male HR
male = sm[sm["sex"] == "m"].sort_values("mean", ascending=False)
print(male[["strain", "mean", "sd", "sem", "nmice", "zscore"]].head(8).to_string(index=False))

# Optional: model-adjusted means (lsmeans) account for covariates fit in MPD's models.
# Use lsmeans when comparing strains across cohorts within a project.
r = requests.get(f"{MPD}/pheno/lsmeans/Jaxwest1", timeout=30).json()
print("LS-mean measures available for Jaxwest1:", r.get("ls_measures", [])[:10])

Module 5: Per-Animal Values — `/pheno/animalvals/{measnum}`

Raw per-animal observations for one measure. Each row carries animal_id, animal_projid, measnum, projsym, sex, stocknum, strain, strainid, value, varname, zscore. Use this for QTL mapping, mixed-effects modeling, or distribution analysis.

import requests, pandas as pd

MPD = "https://phenome.jax.org/api"

# Per-animal heart rate values
r = requests.get(f"{MPD}/pheno/animalvals/15101", timeout=30)
ad = pd.DataFrame(r.json()["animaldata"])
print(f"Animals measured: {len(ad)}")
print(ad[["animal_id", "strain", "sex", "value", "zscore"]].head(8).to_string(index=False))

# Compare male-only strain distributions
male = ad[ad["sex"] == "m"]
stats = male.groupby("strain")["value"].agg(["mean", "std", "count"]).round(2)
print(f"\nMale HR by strain (top 5 by mean):")
print(stats.sort_values("mean", ascending=False).head(5))

Module 6: Ontology-Based Measure Discovery — `/pheno/measures_by_ontology/{ont_term}`

Find every MPD measure annotated to a Mammalian Phenotype (MP), Vertebrate Trait (VT), or Mouse Anatomy (MA) ontology term. Optional this_term_only=yes disables descendant-term expansion; omit_baseline=yes filters out baseline measures; collapse_series=yes collapses repeated time-points.

import requests, pandas as pd

MPD = "https://phenome.jax.org/api"

# MP:0001262 = "decreased body weight"
r = requests.get(f"{MPD}/pheno/measures_by_ontology/MP:0001262",
                 params={"omit_baseline": "yes", "collapse_series": "yes"},
                 timeout=30).json()
print(f"Measures mapped to MP:0001262 ('{r['ontology_terms'][0]['descrip']}'): {r['count']}")
if r["count"]:
    df = pd.DataFrame(r["measures"])
    print(df[["measnum", "varname", "descrip", "projsym"]].head(8).to_string(index=False))
else:
    print("(no direct measures; consider a broader parent term)")

Module 7: Strain Nomenclature — `/straininfo`

Validate and normalise strain names. Accepts name, stocknum, or mginum query params. Returns both jaxinfo[] (JAX availability/nomenclature) and mpdinfo[] (MPD's own metadata, including how many projects test this strain).

import requests

MPD = "https://phenome.jax.org/api"

# Validate C57BL/6J — note `requests` URL-encodes the slash automatically in params
r = requests.get(f"{MPD}/straininfo", params={"name": "C57BL/6J"}, timeout=30).json()
jax = r["jaxinfo"][0]
mpd = r["mpdinfo"][0]
print(f"JAX:  {jax['nomenclature']}  stock={jax['stocknum']}  status={jax['avl_status']}")
print(f"MPD:  longname={mpd['longname']}  type={mpd['straintype']}  "
      f"projects={mpd['nproj']}  snp_projects={mpd['nsnpproj']}  MGI={mpd['mginum']}")

Module 8: Gene Info — `/geneinfo/{symbol}`

Mouse-gene coordinates (GRCm39, in bp), strand, MGI ID, and a short description. Note the response key is the literal string "gene info" (with a space).

import requests

MPD = "https://phenome.jax.org/api"

r = requests.get(f"{MPD}/geneinfo/Lep", timeout=30).json()
for g in r["gene info"]:
    print(f"{g['descrip']}  chr{g['chrom']}:{g['startbp']:,}–{g['endbp']:,} ({g['strand']})")
    print(f"  type: {g['featuretype']}  MGI: {g['mginum']}  cM: {g['centimorgan']}")

Key Concepts

Measnums Are Project-Scoped, Not Global

Every measnum belongs to exactly one project. 15101 is "heart rate" only within Jaxwest1; the same physiological trait in another project has a different measnum (e.g., 35702 for body weight in Lightfoot1). Never hardcode measnums for a trait — always resolve them by:

Finding the project: GET /projects?panelsym=... or GET /projects?investigator=...
Listing its measures: GET /pheno/measureinfo/{projsym}
Picking the right varname / descrip / units triple from the response

Or go the other way and discover measures by ontology term first (Module 6).

Selectors: Projsym vs Measnum

Most /pheno/* endpoints take a "selector" path parameter that's overloaded:

| Endpoint | Accepts as selector |

|----------|---------------------|

| /pheno/strainmeans/{selector} | projsym (e.g., Jaxwest1) or one or more comma-separated measnums |

| /pheno/lsmeans/{selector} | same |

| /pheno/measureinfo/{selector} | same |

| /pheno/animalvals/{measnum} | measnum only (use measureinfo to discover) |

| /pheno/animalvals/series/{measnum} | for timecourse/dose-response series |

If you pass an unrecognised selector you get a 400 JSON response ({"error": "...selector arg must either be measure IDs or a project symbol"}), not an HTML 404 — those are diagnostic and worth surfacing.

Strain Means vs LS-Means

strainmeans are unadjusted: simple per-strain × per-sex arithmetic means of the raw animal values.
lsmeans are model-adjusted least-squares means from MPD's pre-fit ANOVA-style models (accounting for cohort, batch, or covariate effects when present).

For cross-project comparisons or analyses sensitive to batch effects, prefer lsmeans when available. For simple ranking and exploratory work, strainmeans is fine.

MPD Sectors (`mpdsector` filter)

| Sector | Content |

|--------|---------|

| pheno | Standard inbred-strain phenotyping projects |

| qtla | QTL Archive — historical mapping studies with markers |

| komp | Knockout Mouse Project (KOMP / JaxLIMS) data |

| snp | SNP genotype panels (use /snpdata) |

| phenoarchive | Archived legacy phenotype projects |

| onestrain | Single-strain deep phenotyping |

Discover the live list any time with GET /project_filters/mpdsector.

Common Workflows

Workflow 1: Pick a Trait → Find a Project → Plot Strain Means

Goal: From "I want to compare heart rate across inbred strains" → land on real data and produce a ranked barplot.

import requests, pandas as pd, matplotlib.pyplot as plt, time

MPD = "https://phenome.jax.org/api"

# 1) Find candidate projects whose name/description hints at the trait
projects = requests.get(f"{MPD}/projects", timeout=30).json()["projects"]
candidates = [p for p in projects
              if any(kw in (p.get("title") or "").lower()
                     for kw in ["cardiovascular", "heart", "ekg", "ecg"])]
print(f"Candidate cardiovascular projects: {len(candidates)}")
for p in candidates[:5]:
    print(f"  {p['projsym']:<15}  nstrains={str(p.get('nstrains') or '--'):>3}  {(p.get('title') or '')[:60]}")

# 2) Inspect measures for the chosen project
projsym = "Jaxwest1"
mi = requests.get(f"{MPD}/pheno/measureinfo/{projsym}", timeout=30).json()["measures_info"]
hr = next(m for m in mi if m["varname"] == "HR")
print(f"\nPicked: {projsym} measnum={hr['measnum']} varname={hr['varname']} ({hr['descrip']}, {hr['units']})")

# 3) Pull strain means, plot male strains ranked
sm = pd.DataFrame(requests.get(f"{MPD}/pheno/strainmeans/{hr['measnum']}", timeout=30).json()["strainmeans"])
male = sm[sm["sex"] == "m"].sort_values("mean", ascending=False).reset_index(drop=True)

fig, ax = plt.subplots(figsize=(9, 4))
bars = ax.bar(male["strain"], male["mean"], yerr=male["sem"],
              color="#1976D2", capsize=3, edgecolor="white")
ax.bar_label(bars, fmt="%.0f", padding=3, fontsize=8)
ax.set_xlabel("Strain")
ax.set_ylabel(f"Mean {hr['varname']} ({hr['units']})")
ax.set_title(f"{hr['descrip']} by inbred strain (male) — {projsym}")
plt.xticks(rotation=30, ha="right")
plt.tight_layout()
plt.savefig("mpd_strain_means.png", dpi=150, bbox_inches="tight")
print("Saved mpd_strain_means.png")

Workflow 2: Per-Animal Data → R/qtl2 CSV Export

Goal: Pull individual animal observations for a measure and shape them into a phenotype file ready for QTL mapping.

import requests, pandas as pd

MPD = "https://phenome.jax.org/api"

measnum = 15101  # heart rate in Jaxwest1
mi = requests.get(f"{MPD}/pheno/measureinfo/{measnum}", timeout=30).json()["measures_info"][0]
ad = pd.DataFrame(requests.get(f"{MPD}/pheno/animalvals/{measnum}", timeout=30).json()["animaldata"])
print(f"measnum {measnum}: {mi['varname']} ({mi['descrip']}, {mi['units']}) — {len(ad)} animals")

# R/qtl2-ready phenotype CSV: rows = individuals, cols = id + covariates + phenotype
out = ad[["animal_id", "strain", "sex", "value"]].rename(
    columns={"animal_id": "id", "value": mi["varname"]}
)
out.to_csv(f"{measnum}_{mi['varname']}_qtl_pheno.csv", index=False)
print(f"Wrote {measnum}_{mi['varname']}_qtl_pheno.csv  ({len(out)} animals)")
print(out.head().to_string(index=False))

Workflow 3: Multi-Measure Heatmap Across a Strain Panel

Goal: Build a wide-format strain × measure table (z-scored) from one project for comparative visualisation.

import requests, pandas as pd, matplotlib.pyplot as plt

MPD = "https://phenome.jax.org/api"
projsym = "Jaxwest1"

# Pull all strain means for the project in one call
sm = pd.DataFrame(requests.get(f"{MPD}/pheno/strainmeans/{projsym}", timeout=30).json()["strainmeans"])

# Use the precomputed z-scores; pivot to strain × varname (male only for simplicity)
male = sm[sm["sex"] == "m"]
wide = male.pivot_table(index="strain", columns="varname", values="zscore", aggfunc="mean")
print(f"Shape: {wide.shape}  (strains × measures)")

# Subset to a handful of measures with full coverage
keep = wide.dropna(axis=1, thresh=int(0.8 * len(wide))).columns[:10]
wide = wide[keep].dropna()
print(f"After coverage filter: {wide.shape}")

fig, ax = plt.subplots(figsize=(8, max(3, 0.3 * len(wide))))
im = ax.imshow(wide.values, aspect="auto", cmap="RdBu_r", vmin=-2, vmax=2)
ax.set_xticks(range(len(wide.columns)))
ax.set_xticklabels(wide.columns, rotation=45, ha="right", fontsize=8)
ax.set_yticks(range(len(wide.index)))
ax.set_yticklabels(wide.index, fontsize=8)
fig.colorbar(im, ax=ax, label="z-score")
ax.set_title(f"{projsym} — strain × measure z-score heatmap (male)")
plt.tight_layout()
plt.savefig("mpd_strain_measure_heatmap.png", dpi=150, bbox_inches="tight")
print("Saved mpd_strain_measure_heatmap.png")

Key Parameters

|-----------|----------|---------|-----------------|--------|

| csv | /projects, /investigators, /projects/{projsym}/dataset, etc. | no | yes | Return CSV instead of JSON |

| json | /projects/{projsym}/dataset | — | yes | Return JSON instead of default CSV |

| this_term_only | /pheno/measures_by_ontology/{ont_term} | no | yes | Disable descendant-term expansion |

| omit_baseline | /pheno/measures_by_ontology/{ont_term} | no | yes | Drop baseline measures from results |

| collapse_series | /pheno/measures_by_ontology/{ont_term} | no | yes | Collapse timecourse/dose series into single entries |

Best Practices

Always discover measnums via /pheno/measureinfo/{projsym} before querying data. Measnums are project-scoped 5-digit integers; there is no global trait → measnum table. Hardcoding measnums you got from elsewhere will silently 404 or return "no data".

Use plural resource paths. MPD uses /projects, /projects/{projsym}/strains, /projects/{projsym}/dataset — not singular. Old MPD documentation and several third-party wrappers list singular paths that return HTML 404s.

Prefer the project-level dataset CSV for bulk analysis. When you want every animal × every measure for a project, GET /projects/{projsym}/dataset returns a single CSV in one request — much faster than looping animalvals per measnum.

Use lsmeans instead of strainmeans when MPD has fit a model. LS-means adjust for covariates (cohort, age, batch) baked into MPD's project-level statistical models. For comparative ranking across strains within a project, lsmeans is the more honest summary when available (/pheno/lsmeans/{projsym} returns the list of ls_measures).

Validate strain names with /straininfo before assuming a match. MPD uses strict JAX canonical nomenclature; nearby synonyms (B6, C57Bl/6, C57BL/6) won't always resolve. /straininfo?name=... returns both JAX and MPD records and tells you the canonical form.

Be polite — add time.sleep(0.3) in loops. MPD doesn't publish a hard rate limit, but the server runs on shared academic infrastructure. Keep bursts under ~5 req/s.

Common Recipes

Recipe: Find Projects Testing a Trait by Free-Text Search

import requests, pandas as pd

MPD = "https://phenome.jax.org/api"

def find_projects(keyword):
    """Search project titles for a keyword (case-insensitive)."""
    projects = requests.get(f"{MPD}/projects", timeout=30).json()["projects"]
    kw = keyword.lower()
    hits = [p for p in projects if kw in (p.get("title") or "").lower()]
    return pd.DataFrame([{
        "projsym": p["projsym"],
        "panel": p.get("panelsym"),
        "nstrains": p.get("nstrains"),
        "year": p.get("projyear"),
        "title": (p.get("title") or "")[:80],
    } for p in hits])

print(find_projects("glucose").head(10).to_string(index=False))

Recipe: Pull a Whole Project as a DataFrame in One Call

import requests, io, pandas as pd

MPD = "https://phenome.jax.org/api"

def load_project_dataset(projsym):
    """Fetch /projects/{projsym}/dataset CSV directly into a DataFrame."""
    r = requests.get(f"{MPD}/projects/{projsym}/dataset", timeout=120)
    r.raise_for_status()
    return pd.read_csv(io.StringIO(r.text))

df = load_project_dataset("Jaxwest1")
print(f"Jaxwest1: {df.shape[0]} animals × {df.shape[1]} columns")
print("Numeric columns:", df.select_dtypes("number").columns.tolist()[:8])

Recipe: Cross-Strain Comparison from Strain Means

import requests, pandas as pd

MPD = "https://phenome.jax.org/api"

# Multiple measnums in one call (comma-separated selector)
selector = "15101,15102,15103"   # HR, QRS, PR from Jaxwest1
sm = pd.DataFrame(requests.get(f"{MPD}/pheno/strainmeans/{selector}", timeout=30).json()["strainmeans"])
wide = (sm[sm["sex"] == "m"]
        .pivot_table(index="strain", columns="varname", values="mean", aggfunc="mean")
        .round(1))
print(wide.head(8).to_string())

Recipe: Resolve Gene → Coordinates → Adjacent Region

import requests

MPD = "https://phenome.jax.org/api"

def gene_window(symbol, flank_kb=100):
    r = requests.get(f"{MPD}/geneinfo/{symbol}", timeout=30).json()
    if not r.get("gene info"):
        return None
    g = r["gene info"][0]
    return {
        "symbol": symbol,
        "chrom": g["chrom"],
        "start": max(0, g["startbp"] - flank_kb * 1000),
        "stop": g["endbp"] + flank_kb * 1000,
        "mgi": g["mginum"],
        "descrip": g["descrip"],
    }

print(gene_window("Lep", flank_kb=50))
# {'symbol': 'Lep', 'chrom': '6', 'start': 29010220, 'stop': 29123877, 'mgi': 'MGI:104663', ...}

Troubleshooting

| Problem | Cause | Solution |

|---------|-------|----------|

| HTTP 404 with HTML body on /strain/..., /procedure, /pheno/query, /measurement/..., /project/... | These paths don't exist — MPD's real API uses plural resource names and a different layout | Use /projects (plural), /projects/{projsym}/dataset, /pheno/strainmeans/{selector}, /pheno/measureinfo/{selector}, /straininfo |

| HTTP 400 with {"error": "...selector arg must either be measure IDs or a project symbol"} | The path's selector arg got something else (a strain name, a varname, a category) | Resolve the right projsym or measnum first via /projects or /pheno/measureinfo/{projsym} |

| HTTP 404 with {"error": "No strainmeans data found for {selector}"} | The selector is the right kind but has no data (e.g., a measnum from a different project, or a typo) | Confirm the measnum exists via /pheno/measureinfo/{measnum}; check it belongs to the project you think |

| KeyError: 'gene info' when parsing /geneinfo/{sym} | Response key has a literal space: "gene info", not gene_info | Access via r.json()["gene info"] exactly |

| dataset endpoint returns plain text instead of JSON | Default content type is CSV | Pass params={"json": "yes"} to force JSON; or parse the CSV with pd.read_csv(io.StringIO(r.text)) |

| Strain name returns empty mpdinfo from /straininfo | Non-canonical name (e.g., B6, C57Bl/6) | Use exact JAX nomenclature (C57BL/6J); try stocknum= lookup if you have the JAX stock number |

| /pheno/measures_by_ontology/{term} returns count: 0 but the term exists | No direct mappings; the term is too specific | Re-query with the term's parent (the response includes ontology_terms[].parent); or drop this_term_only=yes |

| HTTP 5xx intermittently on large CSV pulls | MPD's per-project datasets can be tens of MB | Increase timeout=120; for very large projects use csv=yes + stream with requests.get(..., stream=True) |

monarch-database — disease-gene-phenotype knowledge graph with HPO ↔ MP cross-mappings; complement MPD's mouse-only data with human disease links
ensembl-database — mouse genome annotation (GRCm39 coordinates, transcripts, VEP) — pairs with MPD /geneinfo for fine-grained gene-model details
gwas-database — human GWAS Catalog SNP-trait associations; conceptual analogue of MPD's QTL projects for human populations
clinvar-database — clinical variant interpretation; relevant when mapping a mouse QTL to a human disease gene

References

Mouse Phenome Database portal — interactive dataset browser and download UI
MPD REST API documentation — authoritative endpoint reference (the one used to build this skill)
Bogue et al., *Mammalian Genome* 2020 — MPD overview paper (architecture, data content, strain coverage)
Jackson Laboratory Inbred Strain Catalog — canonical strain nomenclature and stock numbers
Mammalian Phenotype Ontology — MP term browser (used by /pheno/measures_by_ontology/{term})
MPD Data Use Policy — CC-BY-4.0 license terms and citation requirements

How to use it

Copy the folder

Take jaechang-hits/mouse-phenome-database from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.

Install what it needs

The instructions reference pip. Without those the skill loads but fails at the first command.