본문으로 건너뛰기
cccskills
무료GitHub에 공개

elife-figure-guide

eLife figure preparation: file formats (TIFF/EPS/PDF), striking image requirements (1800x900 px), figure supplement naming, and image screening policy treating selective enhancement as misconduct.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

연동 도구Excel
설치 방법 보기

포함된 파일(1)

  • SKILL.md12.6 KB

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

eLife Figure Preparation Guide

Overview

This guide provides the complete specifications for preparing figures for submission to eLife. eLife is known for its open-access model, figure supplement system, striking image requirements, and a strict image screening policy where selective enhancement of scientific images is treated as research misconduct.

Official reference: https://elife-rp.msubmit.net/html/elife-rp_author_instructions.html


Resolution Requirements

Image TypeMinimum Resolution
Striking images1800 x 900 pixels minimum
Regular figuresNo strict DPI mandate (standard 300 DPI recommended)
from PIL import Image

def check_elife_resolution(image_path, is_striking=False):
    """Check if image meets eLife resolution requirements.

    Args:
        is_striking: True for striking/graphical abstract images
    """
    img = Image.open(image_path)
    w, h = img.size
    dpi = img.info.get('dpi', (72, 72))

    print(f"Dimensions: {w} x {h} px")
    print(f"DPI: {dpi[0]} x {dpi[1]}")

    if is_striking:
        if w >= 1800 and h >= 900:
            print("PASS: Meets striking image minimum (1800x900 px)")
            return True
        else:
            print(f"FAIL: Striking image needs 1800x900 px, got {w}x{h}")
            return False
    else:
        if dpi[0] >= 300:
            print("PASS: Resolution meets standard (300+ DPI)")
        else:
            print(f"NOTE: DPI is {dpi[0]}; 300+ DPI recommended")
        return True

File Format

Regular Figures

FormatAccepted
TIFFYes
EPSYes
PDFYes
JPEG / JPGYes
GIFYes
PS (PostScript)Yes
RTFYes
Microsoft ExcelYes (for data-based figures)
CorelDrawYes

Striking Images

FormatRecommended
PNGYes (preferred)
TIFFYes
JPEGYes

Figure Size and Dimensions

Striking Images (Graphical Abstract)

  • Minimum: 1800 x 900 pixels
  • Format: Landscape orientation
  • Composed of 1-2 panels maximum
  • No labels or text should be included

Regular Figures

  • Each figure uploaded as an individual file
  • No specific column width requirements published

Color Mode

  • No explicit RGB/CMYK mandate in published guidelines
  • Follow standard practice: RGB for digital submission

Font Requirements

Striking Images

  • No labels or text should be included in striking images

Regular Figures

  • No strict font specifications published
  • Follow standard scientific figure conventions (sans-serif, 6-8 pt)
import matplotlib.pyplot as plt

def set_elife_fonts():
    """Configure Matplotlib with standard settings for eLife."""
    plt.rcParams.update({
        'font.family': 'sans-serif',
        'font.sans-serif': ['Helvetica', 'Arial'],
        'font.size': 7,
        'axes.labelsize': 7,
        'axes.titlesize': 8,
        'xtick.labelsize': 6,
        'ytick.labelsize': 6,
        'legend.fontsize': 6,
    })

Labeling Conventions

Figure Supplements (eLife-Specific System)

eLife uses a unique figure supplement system instead of traditional supplementary figures:

  • Format: Figure 1--Figure Supplement 1, Figure 1--Figure Supplement 2, etc.
  • Each supplement should have a short title and an optional legend
  • Supplements are linked to their parent figure in the publication

Striking Images

  • Must NOT contain any labels or text
  • Limited to 1-2 panels
def generate_elife_supplement_names(main_figure_num, n_supplements):
    """Generate eLife figure supplement naming.

    Args:
        main_figure_num: The main figure number (1, 2, 3, ...)
        n_supplements: Number of supplements for this figure

    Returns:
        List of supplement name strings
    """
    names = []
    for i in range(1, n_supplements + 1):
        names.append(f"Figure {main_figure_num}--Figure Supplement {i}")
    return names

# Example
supplements = generate_elife_supplement_names(3, 4)
for s in supplements:
    print(s)
# Output:
# Figure 3--Figure Supplement 1
# Figure 3--Figure Supplement 2
# Figure 3--Figure Supplement 3
# Figure 3--Figure Supplement 4

Image Integrity and Manipulation Policy

CRITICAL: eLife Screens Images Routinely

eLife routinely screens submitted images for inappropriate digital processing. This is not a random check — it is a systematic process.

PROHIBITED

  • Enhancing, obscuring, moving, removing, or introducing any specific feature within an image
  • Selective enhancement constitutes research misconduct
    • Example: Adjusting the intensity of a single band in a blot = misconduct

PERMITTED

  • Brightness, contrast, and color balance adjustments applied equally across the entire image
  • Adjustments must NOT obscure or eliminate information present in the original

Consequences

  • Inappropriate image manipulation may result in:
    • Manuscript rejection
    • Investigation by the author's institution
    • Public correction or retraction of published articles

Python Quick Start: Full Validation

from PIL import Image
import os

def validate_elife_figure(image_path, is_striking=False):
    """Full validation of a figure against eLife requirements."""
    img = Image.open(image_path)
    issues = []
    w, h = img.size

    # 1. Striking image checks
    if is_striking:
        if w < 1800 or h < 900:
            issues.append(f"Striking image: {w}x{h} px below 1800x900 minimum")
        if w < h:
            issues.append("Striking image should be landscape orientation")

    # 2. Resolution check (standard recommendation)
    dpi = img.info.get('dpi', (72, 72))
    if not is_striking and dpi[0] < 300:
        issues.append(f"Resolution {dpi[0]} DPI; 300+ DPI recommended")

    # 3. Format check
    fmt = img.format
    accepted = ['TIFF', 'JPEG', 'PNG', 'EPS', 'PDF', 'GIF']
    if fmt and fmt.upper() not in accepted:
        issues.append(f"Format '{fmt}' may not be standard")

    if is_striking:
        striking_formats = ['PNG', 'TIFF', 'JPEG']
        if fmt and fmt.upper() not in striking_formats:
            issues.append(f"Striking image: prefer PNG, TIFF, or JPEG (got {fmt})")

    # Report
    print(f"=== eLife Figure Validation {'(Striking)' if is_striking else ''} ===")
    print(f"Dimensions: {w} x {h} px")
    print(f"DPI: {dpi[0]} x {dpi[1]}")
    print(f"Color mode: {img.mode}")
    print(f"Format: {fmt}")

    if issues:
        print(f"\nISSUES FOUND ({len(issues)}):")
        for issue in issues:
            print(f"  - {issue}")
    else:
        print("\nAll checks PASSED")

    print("\nWARNING: eLife routinely screens images for manipulation")
    print("Selective enhancement (e.g., single band intensity) = MISCONDUCT")

    return len(issues) == 0

Key Concepts

Figure Supplement System

eLife replaces traditional supplementary figures with a linked figure supplement system. Supplements are named "Figure X--Figure Supplement Y" and are directly associated with their parent figure in the publication. Each supplement requires a short title and optional legend. This system improves discoverability compared to buried supplementary files.

Striking Image Requirements

eLife requires a striking image (graphical abstract) with specific constraints: minimum 1800 x 900 pixels, landscape orientation, 1-2 panels maximum, and no labels or text. This image appears prominently in article listings and social media. It must be a standalone visual that conveys the paper's main finding.

Routine Image Screening Policy

eLife routinely screens all submitted images for inappropriate digital processing. This is systematic, not random. Selective enhancement of any part of a scientific image (e.g., adjusting intensity of a single band in a blot) is treated as research misconduct and may result in rejection, institutional investigation, or retraction.

Decision Framework

What type of figure are you preparing?
├── Striking image (graphical abstract)
│   ├── Dimensions: 1800 x 900 px minimum, landscape
│   ├── Format: PNG (preferred), TIFF, or JPEG
│   └── Content: No text, no labels, 1-2 panels max
├── Main figure
│   ├── Standard format: TIFF, EPS, PDF, JPEG
│   └── Upload as individual file
└── Figure supplement
    ├── Naming: "Figure X--Figure Supplement Y"
    ├── Include short title
    └── Same format requirements as main figures
ScenarioFormatResolutionNotes
Striking imagePNG1800 x 900 px minimumNo text or labels
Western blotTIFF300+ DPIWill be screened for manipulation
Graph or chartEPS or PDFVectorIndividual file upload
MicrographTIFF300+ DPIUniform adjustments only
Supplementary panelSame as mainSame as mainName as "Figure X--Figure Supplement Y"
Data from ExcelExcel (.xlsx)N/AAccepted for data-based figures

Best Practices

  1. Prepare striking images without any text: eLife strictly prohibits labels or text in striking images. The image must communicate visually without words
  2. Use the figure supplement system correctly: Name supplements using the exact format "Figure X--Figure Supplement Y" with double hyphens. Include a short descriptive title for each
  3. Apply all adjustments uniformly to entire images: eLife's screening software detects selective enhancement. Any brightness/contrast change must cover the whole image
  4. Maintain original unprocessed data: Given eLife's screening policy, having originals available demonstrates integrity and speeds up any inquiries
  5. Submit each figure as an individual file: Do not combine multiple figures into a single file. Each figure gets its own upload
  6. Use landscape orientation for striking images: Portrait-oriented striking images do not meet eLife's requirements. Ensure width exceeds height with minimum 1800 x 900 pixels
  7. Describe all processing in Methods: Document every image processing step (software used, adjustments applied) to preempt screening concerns

Common Pitfalls

  1. Adding text or labels to striking images: Any text in the striking image violates eLife requirements
    • How to avoid: Create a purely visual composition; remove all annotations, labels, and text overlays
  2. Selectively enhancing parts of scientific images: Adjusting one region (e.g., a single gel lane) while leaving others unchanged is treated as misconduct
    • How to avoid: Apply all brightness, contrast, and color adjustments to the entire image uniformly
  3. Using incorrect figure supplement naming: Missing the double-hyphen format or incorrect numbering breaks the linking system
    • How to avoid: Use the exact format "Figure X--Figure Supplement Y" with double hyphens (--) and sequential numbering
  4. Submitting striking images below minimum size: Images under 1800 x 900 pixels are rejected
    • How to avoid: Check pixel dimensions before upload; create striking images at or above the minimum
  5. Submitting portrait-oriented striking images: eLife requires landscape orientation for striking images
    • How to avoid: Ensure width (1800+ px) exceeds height (900+ px) in the final composition

Pre-Submission Checklist

Before submitting figures to eLife, verify:

  • Striking image: minimum 1800 x 900 px, landscape, no text/labels
  • Striking image format: PNG, TIFF, or JPEG
  • Regular figures: 300+ DPI recommended
  • File format: TIFF, EPS, PDF, JPEG, or other accepted format
  • Each figure uploaded as individual file
  • Figure supplements named: "Figure X--Figure Supplement Y"
  • Each supplement has a short title
  • No selective enhancement of any image region
  • All brightness/contrast adjustments applied to entire image
  • Adjustments do not obscure or eliminate original information
  • Original unprocessed data retained
  • Aware that eLife routinely screens images for manipulation

References

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

API + Python SDK for ordering cell-free protein expression and binding assays. Submit sequences for expression (10–100 µg), measure binding affinity (KD) against targets, track status, and retrieve results programmatically — no wet-lab setup. Built for ML-guided directed evolution and antibody/nanobody optimization. Requires Adaptyv account and API key.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

jaechang-hits/SciAgent-Skills3812026년 10월 16일 업데이트

aeon

무료

scikit-learn compatible Python toolkit for time series ML: classify, cluster, regress, segment, transform with 30+ algorithms (ROCKET, InceptionTime, KNN-DTW, HIVE-COTE, WEASEL). Handles panel, multivariate, and unequal-length series. Maintained successor to sktime. Alternatives: sktime (larger ecosystem), tslearn (fewer algorithms), catch22 (features only).

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

jaechang-hits/SciAgent-Skills3812026년 10월 16일 업데이트

AiZynthFinder retrosynthetic route planning (CASP) from AstraZeneca Molecular AI. Monte Carlo tree search guided by a template-based neural expansion policy recursively disconnects a target SMILES until precursors are found in a purchasable stock. Covers config.yml (v4 format), aizynthcli batch screening, the AiZynthFinder/AiZynthExpander Python API, one-step disconnections, custom stocks via smiles2stock, scorers, Retro*/breadth-first/DFPN search alternatives, and reading output.json.gz / trees.json. Use for synthesis route planning, synthesizability screening, and building-block/precursor search. For reaction barriers use neb-irc-activation-energy; for 2D reaction scheme drawing use rdkit-chemdraw-cdxml.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

jaechang-hits/SciAgent-Skills3812026년 10월 16일 업데이트

Access AlphaFold DB's 200M+ predicted structures by UniProt ID. Download PDB/mmCIF, analyze pLDDT/PAE, bulk-fetch proteomes via Google Cloud. For experimental structures use PDB; for prediction use ColabFold or ESMFold.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

jaechang-hits/SciAgent-Skills3812026년 10월 16일 업데이트

Annotated matrices for single-cell genomics. Stores X with obs/var metadata, layers, embeddings (obsm/varm), graphs (obsp/varp), uns. Use for .h5ad/.zarr I/O, concatenation, scverse integration. For analysis use scanpy; for probabilistic models use scvi-tools.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

jaechang-hits/SciAgent-Skills3812026년 10월 16일 업데이트

GRN inference from expression via GRNBoost2 (gradient boosting) or GENIE3 (Random Forest). Load matrix, filter by TFs, infer TF-target-importance links, save network. Dask-parallelized to single-cell scale. Core SCENIC component.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

jaechang-hits/SciAgent-Skills3812026년 10월 16일 업데이트

jaechang-hits의 스킬 모두 보기

이 스킬의 문제 신고