Grant Writing

Overview

A grant application is a persuasive, structured document that convinces reviewers a project is significant (it matters), innovative (it is new), and feasible (the team can do it). Funders prescribe the structure, the format, and the review criteria; the agent's job is to write to those constraints. This skill covers the pipeline for three of the largest funders in the world:

• NIH (R01, R21, R03, K awards): Specific Aims (1 page), Research Strategy (Significance, Innovation, Approach; 6/12 pages), Biosketch (NIH biographical sketch, 5 pages), Data Management and Sharing Plan (DMS, 1 page), Budget Justification, Letters of Support.
• NSF (Career, Core, EAGER, RaMP): Project Summary (1 page), Project Description (15 pages including Broader Impacts and Results from Prior NSF Support), Biosketch (2 pages), Data Management Plan, Budget and Budget Justification (5 pages).
• EU Horizon Europe: Concept, Objectives, Methodology (Part B1), Implementation (work packages, deliverables, milestones), Consortium description, Budget (personnel, equipment, travel, other, indirect).

For each, this skill drafts the prose, justifies the budget with Python, queries NIH RePORTER and NSF Award Search for related prior work, and assembles the materials into a single PDF for institutional review.

When to use this skill

• You are writing a new NIH R01/R21/R03/K application.
• You are writing an NSF proposal (CAREER, Core, EAGER, etc.).
• You are writing an EU Horizon Europe proposal (RIA, IA, CSA).
• You need to draft a Specific Aims page or a biosketch.
• You need a Data Management and Sharing (DMS) Plan or DMP.
• You need a budget justification that totals to the cap.
• You are responding to a resubmission (NIH A1) and need to write the introduction and Strengths/Weaknesses response.
• You need Broader Impacts (NSF) or Impact/Exploitation (EU).

Do NOT use this skill for project reports (use scientific-writing) or for conference talks (use scientific-presentations).

Prerequisites

• Python 3.10 or newer.
• Libraries:

  pip install requests jinja2 pyyaml python-docx

• Pandoc and a TeX distribution for PDF assembly: sudo apt install pandoc texlive-full latexmk.
• Institutional Sponsored Programs Office (SPO) to submit through grants.gov / Research.gov / Funding & Tenders Portal. The agent drafts locally; the SPO submits.
• For NIH RePORTER and NSF Award Search no API key is required.

Core workflows

1. Draft Specific Aims (1 page, NIH/NSF-canonical structure)

from jinja2 import Template
template = Template("""
# Specific Aims
{{ hypothesis }}

Aim 1: {{ aim1 }}
Aim 2: {{ aim2 }}
Aim 3: {{ aim3 }}

Expected outcome: {{ outcome }}.
Innovative because {{ innovation }}.
""".strip())
text = template.render(
    hypothesis="...",
    aim1="Determine the role of X in Y using single-cell ...",
    aim2="Test whether inhibiting X reverses Y in a model ...",
    aim3="Validate X as a biomarker in a clinical cohort ...",
    outcome="... mechanistic, preclinical, and translational evidence ...",
    innovation="... integrates spatial transcriptomics with ...")
open("specific_aims.md", "w").write(text)

2. Justify the budget with Python (NIH modular or detailed)

def budget(person_years, salary, fringe=0.29, inflation=0.03, years=5):
    rows = []
    for y in range(years):
        base = salary * (1 + inflation) ** y
        rows.append({"year": y+1, "salary": base,
                     "fringe": base*fringe,
                     "total": base*(1+fringe)})
    return rows
for r in budget(person_years=0.5, salary=125000, fringe=0.29, years=5):
    print(r)
# Add tuition, equipment, travel, materials, publication, indirect (F&A)

3. Biosketch (NSF 2-page / NIH 5-page)

from jinja2 import Template
tpl = Template("""# Biographical Sketch
Name: {{ name }}, {{ title }}
Position: {{ position }}
Education:
{% for e in edu %}- {{ e }}
{% endfor %}

Appointments:
{% for a in appts %}- {{ a }}
{% endfor %}

Products (last 5 years):
{% for p in prods %}- {{ p }}
{% endfor %}

Synergistic Activities:
{% for s in syn %}- {{ s }}
{% endfor %}
""")
open("biosketch.md","w").write(tpl.render(
    name="Ada Lovelace", title="Associate Professor",
    position="Dept. of Computational Biology, Example University",
    edu=["PhD 2014, Stanford University", "BS 2009, MIT"],
    appts=["2020-present: Associate Professor", "2014-2020: Assistant Professor"],
    prods=["Lovelace A. (2024). ...", "Lovelace A., Turing A. (2023). ..."],
    syn=["Organizer, Regional Code Club",
         "Advisory board, Public Data Initiative",
         "Mentor, 4 PhD students"]))

4. Data Management and Sharing Plan (DMS, 1 page)

from jinja2 import Template
dms = Template("""# Data Management and Sharing Plan
## Data types
{{ types }}

## Tools, standards, formats
{{ standards }}

## Storage and access
{{ access }}

## Preservation
{{ preservation }}

## Sharing
{{ sharing }}

## Oversight
{{ oversight }}
""".strip()).render(
    types="Genomic (FASTQ, BAM, VCF), imaging (OME-TIFF), code (Python).",
    standards="Genome Reference Consortium GRCh38, FAIRsharing standards.",
    access="NCBI SRA, BioStudies; institutional AWS S3 during the project.",
    preservation="10 years per NIH policy; Zenodo DOIs for code releases.",
    sharing="Immediate release on publication; controlled-access for PII.",
    oversight="PI and data manager; quarterly FAIR-compliance audits.")
open("dms.md","w").write(dms)

5. Query NIH RePORTER for related prior work

import requests
url = "https://api.reporter.nih.org/v2/projects/search"
payload = {"criteria": {"project_nums":["R01AI123456"],
                         "fiscal_years":[2020,2021,2022,2023,2024]}}
r = requests.post(url, json=payload, timeout=30).json()
for p in r["results"][:5]:
    print(p["project_title"], "|", p["agency_administrator"])

Best practices

• Read the funding opportunity announcement (FOA/NOFO) end to end before drafting; the page limits, review criteria, and required attachments are non-negotiable.
• Lead with a central hypothesis and 2-4 Specific Aims; reviewers remember a one-sentence thesis if it is in the first paragraph.
• Use a Gantt chart or work-package table to show feasibility over the project period; reviewers score feasibility heavily.
• Provide preliminary data for every Aim; if there is no preliminary data, that Aim is at risk of being scored as too risky.
• Address rigor (sample size justification, sex as a biological variable, authentication of biological resources) and reproducibility explicitly in Approach — NIH requires it.
• Write broader impacts (NSF) that are concrete and measured: names of partner schools, expected reach, evaluation plan. Vague "will train graduate students" fails review.
• Show the team can do it: each Aim needs a co-Investigator with preliminary data in that area; no single PI cover-all.
• For resubmissions (NIH A1), write a 1-page Introduction naming each prior weakness, your response, and the page where it is addressed; do not relitigate the original critique.
• Use plain language in the Aims page; reviewers skim in 15 minutes — your Aims page must stand alone.
• Numbers must reconcile: Aims page narrative, budget, and DMS plan must all show the same years, the same data volumes, and the same co-investigators. Reviewers and SPOs catch discrepancies.

Common pitfalls

• Aims that are independent projects rather than coordinated steps toward a single hypothesis.
• Innovation section that lists methods ("we use single-cell RNA-seq") — innovation is the conceptual advance, not the technique.
• Approach written without alternative approaches and pitfalls; reviewers score feasibility higher when risks and mitigations are explicit.
• Budget that asks for 100% effort on a single PI across two large grants — NIH caps at 12 person-months per person across R01s in practice.
• No preliminary data for the most innovative Aim — the riskiest work needs the most preliminary evidence.
• Broader Impacts (NSF) that are generic ("promote diversity") rather than specific ("Partner with X HBCU; co-mentor 2 REU students/year; release all code under MIT").
• Data Management Plan copied from another proposal and not edited for the actual data types in this project — desk-reject trigger.
• Biosketch with >5-year products or >2 pages (NSF); reviewers reject without review.
• Resubmission that ignores the prior critique; NIH asks "what changed?" and reviewers look for the answer in the Introduction.

Worked example

Drafting the Specific Aims and DMS for a fictional NIH R01: "Mechanism of microglial dysfunction in early Alzheimer's disease."

from jinja2 import Template
aims = Template("""
# Specific Aims (R01, 1 page)
Alzheimer's disease (AD) progresses silently for decades before cognitive
decline. Microglia enter a senescent state that may drive this window.
Our **central hypothesis** is that TREM2 loss drives microglial senescence
early in AD, and reactivating TREM2 reverses it.

**Aim 1. Characterize microglial senescence in early AD.** We will
profile 200 postmortem samples (Braak I-III vs controls) with snRNA-seq
and spatial transcriptomics.

**Aim 2. Test causality in humanized mice.** A knock-in mouse carrying
the R47H TREM2 variant will be crossed with App NL-G-F and rescued with
a small-molecule TREM2 agonist.

**Aim 3. Validate biomarkers in CSF.** A 200-subject cohort will be
screened for senescence-derived CSF proteins.

**Expected outcomes**: a mechanistic, preclinical, and translational
chain from gene to biomarker.

**Innovation**: integrates spatial transcriptomics and a first-in-class
TREM2 agonist, in human, mouse, and CSF cohorts.
""".strip())
open("specific_aims.md", "w").write(aims.render())

DMS plan (excerpt):

from jinja2 import Template
dms = Template("""
# Data Management and Sharing Plan
## Data types
snRNA-seq (h5ad, ~5TB), spatial transcriptomics (tiff+json, ~10TB),
CSF proteomics (csv, ~10MB), code (Python).

## Standards
BICAN Cell Registry, HCA Tier 1 Metadata, FAIRsharing.

## Storage & access
Institutional S3 during project; NCBI GEO/SRA on publication; Zenodo for
code releases with a DOI per release.

## Preservation
10 years per NIH policy (NOT-OD-21-013).

## Sharing
Release at time of publication (open access); CSF proteomics released as
controlled access via dbGaP if PII is present.

## Oversight
PI + data manager; quarterly FAIR-compliance audits using FAIRshake.
""".strip())
open("dms.md", "w").write(dms.render())

Assemble with pandoc specific_aims.md dms.md -o application.pdf.

Tools & libraries

References

• NIH Grants & Funding policy library: https://grants.nih.gov/policy/
• NIH DMS Policy (NOT-OD-21-013): https://grants.nih.gov/grants/guide/notice-files/NOT-OD-21-013.html
• NIH Biosketch format page: https://grants.nih.gov/grants/forms/biosketch.htm
• NSF PAPPG (Proposal & Award Policies & Procedures Guide): https://www.nsf.gov/policies/pappg
• EU Horizon Europe Programme Guide: https://ec.europa.eu/info/funding-tenders/opportunities/docs/2021-2027/horizon/guide/programme-guide_horizon_en.pdf
• NIH RePORTER API: https://api.reporter.nih.gov/
• NSF Award Search: https://www.nsf.gov/awardsearch/
• NIAID sample applications & summary statements: https://www.niaid.nih.gov/grants-contracts/sample-applications

Ethics & safety

• Truthful claims. Every preliminary result must be in your lab's notebooks; never invent pilot data, sample sizes, or collaborations.
• Authorship/credit on the team. Use a Key Personnel table with person-months per person; named individuals consent to the effort.
• Conflict of interest. Disclose financial COIs (equity, consultancy, patents) for every key person; NIH and NSF both require this.
• Plagiarism. Do not copy boilerplate from funded proposals without permission; even your own prior boilerplate must be re-cited.
• Rigor and reproducibility. State sample-size justification, sex as biological variable, authentication of cell lines and antibodies, and randomization/blinding.
• Human subjects and animals. Reference the IRB/IACUC approval numbers and the inclusion of women, minorities, and children per NIH policy.
• Data sharing. DMS plans are binding; failure to share can result in funding restrictions on future awards.
• AI use disclosure. Disclose AI assistance in drafting, analysis, or budget assembly per NIH Notice NOT-OD-25-117 (or the current NIH guidance). AI cannot be a co-Investigator or be credited as personnel.