Module 04B · Lesson 04

Communication and Writing — Translating Computational Work for the Rest of the Org

Reading time: 20 minutes Track: Role Path — Computational Biology Prerequisites: Module 04B · Lessons 01-03

What this lesson does

You produce excellent computational work. The rest of your organization makes decisions based on what they understand of your work — which is often a fraction of what you produced.

This lesson addresses the gap. AI is genuinely powerful for translating between technical and non-technical audiences, drafting reports that get read, and producing visualizations that communicate. It's also where you'll be tempted to ship AI-generated writing without enough revision, which is where the failure modes live.

By the end of this lesson, you'll be able to:

Draft reports, summaries, and presentations faster while maintaining accuracy
Translate computational results into language non-computational colleagues can act on
Build presentations that communicate without distorting
Recognize the specific failure modes of AI-assisted communication in computational biology

01 · The translation problem

A persistent challenge in computational biology: your work is precise, technical, and detailed. The people making decisions on your work are often not computational. They want answers, not analyses.

The translation between "what your analysis shows" and "what the decision-maker needs to know" is where computational biologists either add disproportionate value or get marginalized.

Computational biologists who can't translate become invisible. Their work is correct but ignored. The decisions get made on intuition or on a colleague's interpretation of the work. The computational biologist gets reduced to "the analyst" — execution-only, no strategic input.

Computational biologists who translate well become strategic. Their work shapes decisions. They're in the room when strategy is set. They're seen as scientific contributors, not just technicians.

AI dramatically lowers the cost of translation. The bottleneck moves from "I don't have time to communicate this well" to "I'm not putting in the effort to make it clear." That's a much more solvable bottleneck.

02 · The three audiences

Most computational work for biotech ends up communicated to three different audiences. The translation is different for each.

Audience 1 — Other computational scientists

Your peer reviewers. They understand the methods, the conventions, the trade-offs.

They want:

Technical detail
Methodology choices and justifications
Reproducibility information
Limitations and caveats stated honestly

The communication challenge: Avoiding under-statement and over-statement. Be precise.

AI's role: Helps with methods sections, vignettes, technical documentation. Standard scientific writing patterns.

Audience 2 — Domain scientists (biologists, chemists, clinicians)

Your collaborators who use your results to inform their work.

They want:

Bottom line up front
Clear connection to biological question
What's confident vs. tentative
What it means for the next experiment

The communication challenge: Conveying confidence calibration without burying it in caveats. Knowing which technical detail matters and which doesn't.

AI's role: Translation from technical to scientific-but-not-computational language. Drafting structured summaries.

Audience 3 — Leadership and decision-makers

Program leads, executives, board members.

They want:

The decision-relevant takeaway in one sentence
Strategic implications
What you'd do next and why
Confidence and risk

The communication challenge: Resisting the temptation to show all your work. Trust the audience to ask follow-up questions if they need more.

AI's role: Generating executive-level summaries, presentations, slide decks. AI is particularly useful here because the genre is well-defined.

The mistake: producing one report and circulating it to all three audiences. The same content rarely serves all three well. Build different artifacts for different audiences.

03 · The structured report pattern

A common deliverable for computational biologists: a written report on an analysis or project. Here's a structure that works and pairs well with AI assistance.

Structure

# Title — clear, specific, action-oriented when possible

## Executive Summary (1 paragraph, max 200 words)
What was the question? What did we find? What does it mean? What's the next step?

## Key Findings (3-7 bullets)
Each bullet states a specific finding in 1-2 sentences with the key number(s).

## Methods (concise but reproducible)
What data, what analysis, what version of tools. Detail sufficient for another computational scientist to reproduce.

## Results (the body of the report)
The findings explained with supporting figures and tables. Organized by question, not by analysis step.

## Limitations (honest)
What this analysis can't tell us. What assumptions might be wrong. What to be cautious about.

## Recommendations (specific, actionable)
What to do next based on these findings. Linked to specific next experiments or decisions.

## Appendix (for the deep dive)
Detailed methods, supplementary figures, code references, version logs.

The AI-assisted drafting workflow

Step 1 — Outline with bullet points:

You write a 1-2 page outline with the key findings, methodology, limitations, and recommendations as bullets. This is your intellectual content.

Step 2 — AI expansion:

Senior computational biologist drafting a structured report. I'm providing an outline below. Expand the outline into a full report following this structure:

- Executive Summary: 1 paragraph, ~150-200 words
- Key Findings: bullets with specific numbers and clear language
- Methods: ~400 words, sufficient for reproducibility
- Results: full prose discussion, ~1500 words, organized around the questions in the outline
- Limitations: honest assessment, ~300 words
- Recommendations: specific actions, ~300 words

Tone: precise but accessible to biologist colleagues. Use specific numbers from the outline; don't invent values. Where the outline lacks needed information, mark as [MORE DETAIL NEEDED].

[Outline attached]

Step 3 — Heavy revision:

The AI draft is structurally complete but needs:

Your specific voice
Strategic framing that AI can't generate
Tightening of long sections
Sharpening of recommendations

Expect to spend at least as much time revising as the AI saved you in drafting. The net is still a major time savings, but the savings come from acceleration, not replacement.

Step 4 — Verification:

All numbers verified against your underlying analysis
All methods statements accurate
All citations (if any) verified
All recommendations consistent with results

04 · The presentation pattern

Slides are a different communication medium. They're seen, not read. They support a verbal presentation, not replace it.

What AI does well for presentations

Generating outline structures
Drafting slide titles and key bullets
Producing alt-text and accessibility content
Suggesting visualizations
Helping with speaker notes

What AI does poorly for presentations

Visual design (AI suggestions are generally bland)
The actual chart construction (AI-suggested charts often misrepresent data)
The narrative arc that connects slides into a story
The specific rhetoric that lands with your specific audience

The AI-assisted presentation workflow

Step 1 — Define the talk:

Write a paragraph describing:

Audience and their starting knowledge
Time available
The single most important takeaway
The decision or action you want the audience to take

Step 2 — Outline generation:

Senior scientist preparing a presentation. The talk has the following parameters:

Audience: [description]
Duration: [N] minutes
Setting: [conference, internal meeting, board, etc.]
Key takeaway: [single sentence]
Desired action: [what you want them to do/think]

Generate a slide-by-slide outline. For each slide, provide:
- Slide title
- 2-3 key bullets or content elements
- Speaker notes (1-2 sentences on what you'll say)
- Visual recommendation (chart, table, image, text)

Aim for [N/2 to N] slides total (one minute per slide is a reasonable pace).

Step 3 — Slide-by-slide content:

For each slide, you populate the actual content. AI helps with:

Bullet phrasing
Chart description (you make the chart)
Speaker notes

Step 4 — Iterate ruthlessly:

The first AI-assisted draft of a presentation is usually too dense. Cut. Cut more. The audience can only absorb so much in your allotted time. Aim for fewer slides with clearer points rather than more slides with everything.

Step 5 — Visual polish:

AI doesn't make slides beautiful. That's your job (or a designer's, if you have one). But the structure and content are 80% there.

05 · The plain-language summary

A specific genre worth its own treatment: summarizing complex computational work for non-computational colleagues.

The pattern

You've completed an analysis. You need to send a 1-paragraph summary to a colleague or in a Slack channel.

Senior scientist writing a plain-language summary. Take the following technical content and produce a single paragraph (~150 words) summary suitable for [audience description].

Requirements:
- Lead with the bottom line in the first sentence
- One additional sentence on what we found
- One sentence on what it means
- One sentence on next step (if relevant)
- Use specific numbers where they matter; avoid technical jargon
- Tone: clear, direct, no hedging beyond what's truly warranted

[Technical content]

The output is usable with light revision. This is one of the highest-ROI uses of AI in computational biology — converting your work into the form colleagues will actually read.

A specific failure mode

AI-generated plain-language summaries can be too uniformly confident or too uniformly cautious. They smooth out the nuance.

The fix: Read the AI summary against your underlying confidence. If you're 90% confident in the main finding but 50% confident in a particular interpretation, the summary should reflect that. AI summaries default to flat confidence.

06 · Talking to non-computational colleagues live

A scenario you'll encounter regularly: a colleague stops by, asks about your latest analysis, and wants a quick update.

AI doesn't help in real-time conversation. But the prep you do before the conversation matters.

The pre-conversation prep

If you know a difficult conversation is coming (a program meeting, a 1:1 with a senior leader), prep by:

Writing a one-sentence summary of what you found
Listing three things they'll likely ask
Drafting one-sentence answers to each
Identifying the one thing you most want them to take away

You can do this with AI assistance:

Senior computational scientist preparing for a 30-minute discussion with [audience]. The topic is [analysis]. The key findings are [bullets].

Generate:
1. The single sentence I should lead with
2. The 3-5 questions this audience is most likely to ask
3. A concise answer to each question
4. The one thing I most want them to take away

You'll find that this prep makes you significantly more effective in the actual conversation. The AI helped you think; the conversation is yours.

07 · Visualization — where AI is more useful than people think

A specific area: making figures that communicate.

AI is not particularly good at visual aesthetics. It is good at:

Suggesting what kind of chart fits which data and message
Generating the code to produce the chart
Critiquing existing charts for clarity and accuracy
Translating between chart types

The figure-critique pattern

Senior scientific visualization expert. Review the following figure for:

1. Does it accurately represent the underlying data?
2. Does it clearly communicate the intended message?
3. Are there design choices that could mislead (axis scaling, color choice, aspect ratio)?
4. What would you change to improve it?
5. For a different audience [specify], would you change the figure significantly?

[Description of figure, or paste the figure if multimodal AI]

The critique is generally substantive. Some suggestions are useful; some are reaches. Use as a checklist of considerations.

The "wrong chart" diagnosis

A common situation: you have data and an instinct that it would make a chart, but you're not sure which chart.

I have the following data: [description]. I want to communicate [message] to [audience]. Suggest 3 different chart types that could work. For each, explain when it would be the best choice and what trade-offs to consider.

The response gives you structured options. You then pick one and have AI write the code.

This works particularly well when you're stuck. Sometimes you're trying to use a bar chart for something that wants to be a scatter plot, and AI surfaces that immediately.

08 · The "what would I say if asked" technique

A specific tool for preparing computational work for review.

After you've drafted a report or presentation, before circulating it, ask:

Senior reviewer of the following analysis. For each of the following audience types, list 3-5 questions you'd ask if presented with this work:

1. A skeptical biostatistician
2. A program leader who wants to make a decision
3. An executive thinking about resource allocation
4. A regulator or external reviewer (if applicable)

For each question, briefly note what a strong response would look like.

[Your work attached]

This produces a structured Q&A document that helps you:

Anticipate the actual questions you'll get
Identify weaknesses in your work before they're surfaced
Sharpen your prepared answers

The Q&A document is also a deliverable in its own right — for important presentations, you can include this as preparation material.

09 · A worked example

A realistic communication scenario.

Setting: You've completed a single-cell RNA-seq analysis of a target tissue. You identified 4 cell populations of interest, with one showing a specific signature that aligns with your team's hypothesis. You need to:

Present to the program team next week (30 minutes)
Send a quick Slack summary to the team lead today
Write a brief report for the broader team this week

Slack summary (drafted in 5 minutes with AI assistance)

Single-cell analysis is complete. Found 4 distinct populations in [tissue], with [population X] showing the expected [marker] signature we predicted. Effect is consistent across all 6 samples. Two populations (A and C) showed unexpected expression patterns that I'd flag for follow-up. Full report by Friday; happy to discuss before then if useful.

Brief report (drafted in 90 minutes, half AI-assisted)

Following the structure from Section 03. Outline written by you; expansion drafted by AI; revision and verification by you.

Presentation (drafted in 3 hours, half AI-assisted)

Outline generation, slide titles, and speaker notes assisted by AI. Visual content and narrative arc by you. Total prep time including practice: ~5 hours.

Without AI: this same set of artifacts would have taken ~2-3 days. The savings come from accelerated drafting, not from skipping any steps.

10 · Knowledge check

Three questions.

Q1. Why is "same report for all audiences" specifically called out as a mistake?

a) It's lazy b) Different audiences (peers, domain scientists, leadership) want different content at different depth — one report rarely serves all three well; build different artifacts for different audiences c) It's against company policy d) Reports are obsolete; only presentations matter

Q2. What's the most accurate description of AI's role in plain-language summaries?

a) AI produces ready-to-send summaries; minimal review needed b) AI accelerates drafting but tends to flatten confidence — read against your actual confidence and revise where the summary is too uniformly confident or too uniformly hedged c) AI can't write plain-language summaries d) Plain-language summaries should always be written manually

Q3. Why does this lesson argue that computational biologists who don't communicate well get marginalized?

a) They're not technical enough b) Decisions get made on intuition or on a colleague's interpretation; the computational biologist gets reduced to execution-only and loses strategic input — translation is what determines whether the work shapes decisions or gets ignored c) Computational work is becoming obsolete d) Communication is more important than analysis

Answers: Q1: b · Q2: b · Q3: b

11 · What's next

Lesson 05 of Module 04B — Capstone playbook. Pull everything together into a personal operating system.

End of Lesson 04.