本文へ移動
cccskills
無料GitHub で公開

lancet-figure-guide

The Lancet figure preparation: resolution (300+ DPI at 120%), preferred editable formats (PowerPoint/Word/SVG), column widths (75/154 mm), Times New Roman, in-house redraw policy.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md11.1 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

The Lancet Figure Preparation Guide

Overview

This guide provides the complete specifications for preparing figures for submission to The Lancet and Lancet family journals. A key distinguishing feature of The Lancet is that most figures are redrawn by in-house illustrators into Lancet house style. Therefore, editable source formats (PowerPoint, Word, SVG) are strongly preferred over rasterized images.

Official reference: https://www.thelancet.com/submission-guidelines


Resolution Requirements

RequirementSpecification
All photographic images300 DPI minimum
Submission size120% of intended publication size

TIP: Supply images 20% larger than publication size to ensure quality is maintained after any resizing by the production team.

from PIL import Image

def check_lancet_resolution(image_path):
    """Check if image meets Lancet resolution requirements."""
    img = Image.open(image_path)
    dpi = img.info.get('dpi', (72, 72))

    print(f"DPI: {dpi[0]} x {dpi[1]}")
    print(f"Minimum required: 300 DPI")
    print(f"TIP: Submit at 120% of publication size")

    if dpi[0] >= 300:
        print("PASS: Resolution meets Lancet requirements")
    else:
        print("FAIL: Need at least 300 DPI")

    return dpi[0] >= 300

File Format

Preferred Formats (Editable)

FormatNotes
PowerPoint (.pptx)Most preferred — easy for in-house redraw
Word (.docx)Accepted for simple figures
SVGScalable vector graphics

Also Accepted

FormatNotes
TIFFFor photographic images
JPEGAcceptable quality
EPSVector format
PDFGeneral purpose

Why Editable Formats?

The Lancet's in-house illustration team redraws most submitted figures into the Lancet house style. Providing editable source files (especially PowerPoint) makes this process smoother and more accurate.


Figure Size and Dimensions

LayoutWidth
1 column75 mm (2.95 in)
2 columns154 mm (6.06 in)
Minimum width107 mm (4.21 in)
  • All panels must fit on a single page
import matplotlib.pyplot as plt

LANCET_WIDTHS = {
    'single':  75 / 25.4,    # 2.95 inches
    'minimum': 107 / 25.4,   # 4.21 inches
    'double':  154 / 25.4,   # 6.06 inches
}

def create_lancet_figure(layout='single', aspect_ratio=0.75):
    """Create a Matplotlib figure sized for The Lancet."""
    width = LANCET_WIDTHS[layout]
    height = width * aspect_ratio

    fig, ax = plt.subplots(figsize=(width, height))
    fig.set_dpi(300)
    return fig, ax

Color Mode

PurposeColor Mode
Online publicationRGB
Print publicationCMYK
  • Color and black-and-white photographs both accepted

Font Requirements

ElementFontSizeStyle
Main figure headingTimes New Roman10 ptBold
Figure legendTimes New Roman10 ptRegular, single-spaced

Key Difference from Other Journals

  • The Lancet uses Times New Roman (serif font), unlike most other biomedical journals that prefer sans-serif fonts
import matplotlib.pyplot as plt

def set_lancet_fonts():
    """Configure Matplotlib for Lancet figure fonts."""
    plt.rcParams.update({
        'font.family': 'serif',
        'font.serif': ['Times New Roman', 'Times'],
        'font.size': 10,
        'axes.labelsize': 10,
        'axes.titlesize': 10,
        'axes.titleweight': 'bold',
        'xtick.labelsize': 8,
        'ytick.labelsize': 8,
        'legend.fontsize': 8,
    })

Labeling Conventions

Rules

  • No box outlines around graphs
  • No titles inside graphs or artwork — include in caption instead
  • Number figures in order of citation in text

Best Practices

  • Keep figures as simple and clear as possible
  • The Lancet illustrators will redraw most figures, so focus on conveying data accurately rather than visual polish

Image Integrity and Manipulation Policy

  • Follow standard biomedical publishing ethics
  • All image adjustments should be applied uniformly
  • Describe any image processing in Methods
  • Maintain original data for potential review

Python Quick Start: Full Validation

from PIL import Image
import os

def validate_lancet_figure(image_path, layout='single'):
    """Full validation of a figure against Lancet requirements."""
    img = Image.open(image_path)
    issues = []

    # 1. Resolution check (300 DPI at 120% size)
    dpi = img.info.get('dpi', (72, 72))
    if dpi[0] < 300:
        issues.append(f"Resolution {dpi[0]} DPI below 300 DPI minimum")

    # 2. Dimension check
    widths_mm = {'single': 75, 'minimum': 107, 'double': 154}
    if layout in widths_mm and dpi[0] > 0:
        actual_w_mm = (img.size[0] / dpi[0]) * 25.4
        target = widths_mm[layout]
        # Should be 120% of publication size
        target_120 = target * 1.2
        print(f"Width: {actual_w_mm:.1f} mm (target at 120%: {target_120:.1f} mm)")

    # 3. Format recommendation
    fmt = img.format
    if fmt and fmt.upper() in ('TIFF', 'JPEG', 'PNG'):
        print(f"NOTE: Format is {fmt}. Lancet prefers PowerPoint/Word/SVG for redrawing")

    # Report
    print(f"=== Lancet Figure Validation ===")
    print(f"Dimensions: {img.size[0]} x {img.size[1]} px")
    print(f"DPI: {dpi[0]} x {dpi[1]}")
    print(f"Color mode: {img.mode}")

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

    return len(issues) == 0

Key Concepts

In-House Figure Redraw Policy

The Lancet's most distinctive feature is that most submitted figures are redrawn by in-house illustrators into Lancet house style. This means editable source formats (PowerPoint, Word, SVG) are strongly preferred over polished raster images. Authors should focus on data accuracy rather than visual refinement.

The 120% Size Rule

The Lancet requires photographic images to be submitted at 120% of intended publication size at 300+ DPI. This buffer ensures quality is maintained after any resizing during production. A figure intended for single-column (75 mm) should be supplied at 90 mm width.

Serif Font Convention

Unlike most biomedical journals that use sans-serif fonts, The Lancet uses Times New Roman as its standard figure font. Headings are 10 pt bold; legend text is 10 pt regular single-spaced. This aligns with Lancet's traditional typographic style.

Decision Framework

What type of figure are you preparing?
├── Data visualization (graph, chart)
│   ├── PowerPoint available → Submit .pptx (preferred for redraw)
│   └── Only image available → SVG or EPS (vector, editable)
├── Photograph or clinical image
│   ├── High quality needed → TIFF at 300+ DPI, 120% size
│   └── Standard quality → JPEG at 300+ DPI, 120% size
└── Schematic or diagram
    ├── Editable source → PowerPoint or Word
    └── Vector only → SVG or EPS
ScenarioFormatResolutionNotes
Bar chart from clinical trialPowerPoint (.pptx)N/A (vector)Lancet redraws in house style
Kaplan-Meier curvePowerPoint or SVGN/A (vector)Provide raw data if possible
Histology micrographTIFF300+ DPI at 120% sizeCannot be redrawn
Flowchart (CONSORT)PowerPoint or WordN/A (editable)Lancet will redraw
Forest plot (meta-analysis)PowerPoint or SVGN/A (vector)Data accuracy is priority

Best Practices

  1. Submit editable formats whenever possible: PowerPoint is most preferred because Lancet illustrators redraw figures. Provide the editable source, not a rasterized export
  2. Supply images at 120% of publication size: Calculate target size (75 mm or 154 mm column width), then multiply by 1.2 for the submission dimensions
  3. Use Times New Roman consistently: The Lancet's serif font convention differs from most journals. Set Times New Roman as default before creating figures
  4. Remove box outlines from graphs: Lancet style does not use borders around graph panels. Remove any automatic axis boxes
  5. Place titles in captions only: Do not embed titles inside figures or graphs. All descriptive text goes in the figure caption
  6. Focus on data accuracy over aesthetics: Since figures will be redrawn, ensure data values are correct and clearly labeled rather than investing in visual polish
  7. Number figures in citation order: Figures must be numbered sequentially in the order they are first cited in the text

Common Pitfalls

  1. Submitting only rasterized images for data figures: Lancet illustrators cannot redraw from JPEG/PNG rasters of graphs
    • How to avoid: Always provide the editable source file (PowerPoint, Word, or SVG) alongside any raster export
  2. Forgetting the 120% size requirement: Submitting at 100% publication size means the image will be undersized after production resizing
    • How to avoid: Multiply the target column width by 1.2 before setting your canvas size
  3. Using sans-serif fonts: Most journals use Arial/Helvetica, but Lancet uses Times New Roman. Sans-serif fonts signal unfamiliarity with Lancet style
    • How to avoid: Set Times New Roman as the default font before creating any Lancet figures
  4. Including box outlines around graphs: Lancet style prohibits boxes around graph panels, a common default in plotting software
    • How to avoid: Disable axis box/frame in your plotting settings (e.g., ax.spines in matplotlib)
  5. Over-polishing figure aesthetics: Since Lancet redraws most figures, spending excessive time on visual refinement is wasted effort
    • How to avoid: Ensure data accuracy and clear labeling; leave visual styling to Lancet's illustration team

Pre-Submission Checklist

Before submitting figures to The Lancet, verify:

  • Photographic images at 300+ DPI
  • Images supplied at 120% of publication size
  • Preferred format: PowerPoint, Word, or SVG (for in-house redraw)
  • TIFF/JPEG/EPS acceptable for photographs
  • 1-column figures: 75 mm wide
  • 2-column figures: 154 mm wide
  • All panels fit on a single page
  • Font is Times New Roman (10 pt bold for headings)
  • No box outlines around graphs
  • No titles inside graphs (use captions)
  • Figures numbered in order of text citation
  • Data clearly and accurately represented (Lancet will redraw)
  • Original data retained for review

References

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

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.

日本語の概要は準備中です。原文の説明を表示しています。

jaechang-hits/SciAgent-Skills3762026年9月29日 更新

aeon

無料

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).

日本語の概要は準備中です。原文の説明を表示しています。

jaechang-hits/SciAgent-Skills3762026年9月29日 更新

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.

日本語の概要は準備中です。原文の説明を表示しています。

jaechang-hits/SciAgent-Skills3762026年9月29日 更新

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.

日本語の概要は準備中です。原文の説明を表示しています。

jaechang-hits/SciAgent-Skills3762026年9月29日 更新

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.

日本語の概要は準備中です。原文の説明を表示しています。

jaechang-hits/SciAgent-Skills3762026年9月29日 更新

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.

日本語の概要は準備中です。原文の説明を表示しています。

jaechang-hits/SciAgent-Skills3762026年9月29日 更新

jaechang-hits のスキルをすべて見る

このスキルの問題を報告する