Wildtype One

Your one trait that AI can’t replace yet “Computers are useless. They can only give you answers.

AI still struggles to replace researchers because of one trait that machines still lack

Researchers who use it consistently reach world-class levels 🏆

What is this trait?

Read today's Wildtype Weekly to find out (in 3 minutes or less 👇)

Love the first line in your "Method" section. We just posted about it.

Also great mechanistic data—congrats to you and the team @martingarridorc.bsky.social

Science thrives on rigor

It is painful in the beginning

But it pays in precision and good data

And of course: reputation in research

— Wildtype One 🧬

Regular testing brings that number down significantly

Regular

Not only when you suspect contamination

Do it regularly

Then add it to your methods section

(4/4)

Regular Mycoplasma testing is such an underrated and underused protocol

15-35% of all cell lines are mycoplasma positive (ATCC)

This means for every 3-7 figures you produce with untested cell lines

One can be myco-compromised

(3/4)

It is our job as researchers to not only show data

But to show their credibility

(2/4)

While irreproducibility is undermining the biggest research breakthroughs

We're hearing more false data today than ever

The literature is being taken with a pinch of salt

This one phrase in the methods section gave a confidence boost for the whole article

(1/4)

🧫 Join 1,000+ researchers getting weekly lab hacks and productivity tools (it’s free) 👉 wildtypeone.substack.com/about

The human brain still wins

— Wildtype One 🧬

(13/13)

What AI has NOT fully solved yet?

Still unstable or hype-heavy:

1/ De novo drug design (partially useful, not reliable)
2/ Hypothesis generation
3/ Causal biological reasoning
4/ Grant writing (assistive only)
5/ Experimental design automation
6/ Multi-omic integration (still messy)

(12/13)

6/ In sequencing and imaging, AI now

- Flags low-quality sequencing runs
- Detects batch effects
- Spots imaging artifacts

Many platforms integrate ML-based QC invisibly

Researchers often don’t realize AI is embedded

(11/13)

5/ Lab automation scheduling & robotics optimization is real

AI scheduling now:

— Optimizes liquid handler timing
— Reduces collision errors
— Improves throughput

Especially in screening facilities and biotech labs

Academia is slower here, but industry has largely adopted it.

(10/13)

Cryo-EM resolution revolution is partially AI-driven

This is completely normalized in structural biology:

No one manually picks thousands of particles anymore.

(9/13)

4/ Cryo-EM particle picking was underappreciated

Particle picking before was manual and exhausting

And it varied among users

Now, deep learning pickers:

— Detect particles reliably
— Dramatically accelerate reconstruction

(8/13)

Human cognition can’t handle 30D space
ML outperforms manual inspection

The hidden truth is that most labs still manually “clean up” ML clusters and use AI as assistant, not authority

But it’s now expected in high-dimensional studies.

(7/13)

3/ Manual gating in flow cytometry is subjective, operator-dependent, and difficult in high-dimensional panels

Machine-learning clustering:

— Identifies populations unbiased
— Handles >20 markers
— Reduces variability

(6/13)

They train a model once and run batch analysis

It cuts time and works across experiments

There are still hidden issue, like training bias, and most labs do not properly validate segmentation accuracy

But culturally, this is now an accepted standard.

(5/13)

2/ Many labs no longer train students to manually count and segment cells

Manual counting and segmentation were always slow, biased, and inconsistent across users

AI cell segmentation now:

— Outperforms humans in many contexts
— Handles crowded images
— Handles 3D stacks

(4/13)

Researchers routinely:

— Map mutations
— Infer binding pockets
— Design truncations
— Identify disordered regions

There are limitations still, but for 70% of basic lab questions? It’s “good enough.”

(3/13)

1/ Protein structure prediction has always been slow

For decades,

no crystal structure
no cryo-EM
no homology model

= no structural hypothesis

This required years + collaborators

Now, you paste the sequence on AlphaFold and get the structure in hours — it's now molecular biology routine

(2/13)

Bench problems AI already solved—and what AI still cannot do 🥼

(a thread)

This is an important reminder that α-synuclein strain diversity can arise intrinsically and that stochastic variability can sway disease phenotypes

Synucleinopathy research needs careful interpretation of PFF-based models

Congrats @raphaellaso.bsky.social and thanks for sharing.

It's fascinating how tissues can maintain function without canonical stem cell-driven regeneration

Continuous endoreplication highlights alternative homeostasis tactics—aging and tissue resilience might really go beyond proliferative renewal

Congrats @benforward3.bsky.social and thanks for sharing

Linking cytoskeletal organization + junctional anchoring via CCM1 + multiscale modeling is a strong model of how endothelial contraction changes vessel diameter

Love the actin dynamics videos as well

Congrats @phnglab.bsky.social and thanks for sharing.

Moving from rabbit epithelial cells to human fibroblast-based cultivation gets us closer to accurate host–pathogen models

A more humanized system should substantially improve mechanistic studies of syphilis pathogenesis and therapy development

Thanks for sharing @bstevensonlab.bsky.social

Love the integration of Hi-C, ChIP-seq, and in vivo validation

Thsi is strong evidence that spatial genome organization plays an active, instructive role in developmental fate decisions

Congrats @jrotwitguez.bsky.social and thanks for sharing.

Measuring the effect of complex metabolite mixtures directly in primary patient cells bridges a long-standing gap between product discovery and clinical relevance

Single-cell resolution adds mechanistic clarity that’s often missing in bulk

Thanks for sharing @brianobachmann.bsky.social

A sharp mechanistic finding—separating thermogenic control from glucose tolerance might tell us why metabolic outcomes dissociate in physio- and pathological states

A much-needed precision to how therapies target brown fat

Congrats @zeltserlab.bsky.social and thanks @adlunglab.com for sharing

Love the first sentence of the Method section: "All cell lines used in this study were verified to be negative for Mycoplasma contamination."

More papers should mention this (and actually test it).

Congrats to your team @saezlab.bsky.social and thanks for sharing.

Such a difficult space... strong expression doesn’t always translate into functional improvement within short follow-up windows

Transparency around safety signals and trial design limitations will inform future gene therapy studies

Thanks for sharing @oligogirl.bsky.social