Ming "Tommy" Tang
@tangming2005
Followers
41K
Following
32K
Media
2K
Statuses
68K
Director of bioinformatics at AstraZeneca. YouTube at chatomics. On my way to helping 1 million people learn bioinformatics. Also talks about leadership.
Boston, MA
Joined December 2011
The guide I wish I had 12 years ago: a step-by-step guide to replicate a genomics paper figure
divingintogeneticsandgenomics.kit.com
8
88
541
Predicting drug responses of unseen cell types through transfer learning with foundation models https://t.co/jvza2nbpA9
0
7
59
SSK—the first U.S.-listed ETF that offers exposure to spot Solana and Solana staking rewards. SSK seeks to stake all (Solana) holdings on the Solana network. Any rewards earned may be distributed to shareholders, via monthly distributions. Distributions are not guaranteed.
1
1
27
after 2 years of work, my weekly chatomics bioinformatics newsletter has grown to almost 10K subscribers. Subscribe and share with your friends to receive weekly dose of bioinformatics tips earned the hard way :) https://t.co/q70UytZ2C4
0
3
3
Predicting drug responses of unseen cell types through transfer learning with foundation models https://t.co/jvza2nbpA9
0
7
59
after 2 years of work, my weekly chatomics bioinformatics newsletter has grown to almost 10K subscribers. Subscribe and share with your friends to receive weekly dose of bioinformatics tips earned the hard way :) https://t.co/q70UytZ2C4
0
3
3
I hope you've found this post helpful. Follow me for more. Subscribe to my FREE newsletter chatomics to learn bioinformatics
divingintogeneticsandgenomics.kit.com
Why Subscribe?✅ Curated by Tommy Tang, a Director of Bioinformatics with 100K+ followers across LinkedIn, X, and YouTube✅ No fluff—just deep insights and working code examples✅ Trusted by grad...
1/ You used autoplot() for PCA and your PC1 and PC2 axes look weird? Don’t blame your data—blame blind trust. Let me explain.
0
0
3
10/ Use packages. But never stop thinking. Understanding beats automation. Every line of code tells a story—make sure it’s the right one.
1
0
0
🎙️ Why talk about "AI Employees" instead of agents? Maybe the paradigm of employee feedback and performance management is the future of agent observability and evaluation... On our latest episode of Deployed we talk with @surojit, founder of @Ema_Unlimited, about his lessons
2
0
14
9/ Key takeaways: autoplot() rescales PCs by default Always read function docs Trust your intuition when things feel off Use raw PCA scores for full control if you want to scale the PCs, you can do it manually too.
1
0
0
8/ This is why bioinformatics isn’t just pushing buttons. It’s thinking through every choice—especially defaults.
1
0
0
7/ If you know ggplot2, you don’t need autoplot() to plot PCA. Direct control = better insight = fewer surprises.
1
0
0
6/ What I learned: if you're confused by a plot, dig deeper. Understand what the function is doing under the hood. Code is trust.
1
0
0
We don’t need new law to build America’s AI bench. We need courage, MOUs, and a 90-day clock.
0
8
65
5/ Want raw scores? Use base prcomp() and ggplot2 directly: pca <- prcomp(iris[,1:4], scale = TRUE) raw_scores <- pca$x[, 1:2] ggplot(data.frame(raw_scores), aes(PC1, PC2)) + geom_point()
1
0
0
4/ This scale shift isn’t a bug—it’s a choice. But it can distort interpretation, especially if you're used to raw PCA plots.
1
0
0
3/ By default, autoplot() rescales PC scores: PCi' = PCi / sd(PCi) * sqrt(n − 1) So your plot is not using the raw PCA values.
1
0
0
2/ Packages are great. But defaults hide a lot. ggfortify::autoplot() scales PCA scores. That’s why your PCs don’t match intuition.
1
0
0
Always curious and full of energy, Beagles turn every walk into an adventure 🐾🎉.
146
607
9K
1/ You used autoplot() for PCA and your PC1 and PC2 axes look weird? Don’t blame your data—blame blind trust. Let me explain.
1
2
15
Enjoy this tweet? follow me @tangming2005 and join my newsletter to learn computational biology
0
0
0
I hope you've found this post helpful. Follow me for more. Subscribe to my FREE newsletter chatomics to learn bioinformatics
divingintogeneticsandgenomics.kit.com
Why Subscribe?✅ Curated by Tommy Tang, a Director of Bioinformatics with 100K+ followers across LinkedIn, X, and YouTube✅ No fluff—just deep insights and working code examples✅ Trusted by grad...
1/ “Deep neural networks (DNN) are just glorified linear models.” You’ve probably heard this. But let’s be honest: it’s both true… and completely wrong.
0
0
0
want to learn deep learning? Watch StatQuest to get started
statquest.org
An epic journey through statistics and machine learning
1
0
0
12/ Don’t dismiss DNNs as just “stacked regressions.” They’re not perfect. But when used right, they’re not glorified linear models. They’re transformation engines.
1
0
0
Who is Zhang Shengmin, China's new CMC vice chair? ✍️ @redharryng @imtienyikrislih
3
6
64