Prof. Nikolai Slavov

@slavov_n

Mentor, scientist & engineer. Director of @ParallelSqTech. Having fun in @SlavovLab with single-cell proteomics & ribosomes. Organizer of @SCP_meeting

Cambridge, USA

Joined April 2014

247Following

19KFollowers

Pinned

Prof. Nikolai Slavov@slavov_n · Feb 15

What determines protein levels ? For many proteins, abundance variation is dominated by clearance (degradation & dilution) variation. Globally, clearance variation across tissues (e.g., brain & pancreas) explains ~ 40 % of protein abundance variation. biorxiv.org/content/10.110…

slavov_n's tweet image. What determines protein levels ?

For many proteins, abundance variation is dominated by clearance (degradation &amp; dilution) variation.

Globally, clearance variation across tissues (e.g., brain &amp; pancreas) explains ~ 40 % of protein abundance variation.

biorxiv.org/content/10.110…

294

131

34.0K

Prof. Nikolai Slavov@slavov_n · Jul 21

AI tools can be very useful for science. Their indiscriminate, haphazard applications also risks producing more but understanding less.

slavov_n's tweet image. AI tools can be very useful for science.

Their indiscriminate, haphazard applications also risks producing more but understanding less.

3.0K

Prof. Nikolai Slavov@slavov_n · Jul 20

I hear about many recruiting opportunities for single-cell proteomics. The opportunities are more than the qualified applicants.

968

Prof. Nikolai Slavov@slavov_n · Jul 20

When doing research, outsourcing something we don’t understand deeply is dangerous. Such outsourcing fundamentally undermines confidence in the conclusions.

1.0K

Prof. Nikolai Slavov@slavov_n · Jul 20

Eukaryotic ribosome biogenesis youtube.com/watch?v=ETkwf6…

207

10.0K

Prof. Nikolai Slavov@slavov_n · Jul 19

More advanced AI chatbots are more likely to oversimplify complex scientific findings. livescience.com/technology/art…

slavov_n's tweet card. More advanced AI chatbots are more likely to oversimplify complex scientific findings based on the way they interpret the data they are trained on, a new study suggests.

983

Prof. Nikolai Slavov@slavov_n · Jul 19

Data leakage is a common problem in AI. It is analogous to ‘teaching to the test’ and leads to overestimation of model capabilities. This article reports that leakage affected 294 papers across 17 scientific fields.

slavov_n's tweet image. Data leakage is a common problem in AI.

It is analogous to ‘teaching to the test’ and leads to overestimation of model capabilities.

This article reports that leakage affected 294 papers across 17 scientific fields.

202

113

16.0K

Prof. Nikolai Slavov@slavov_n · Jul 19

Predictions apply to one task. World models generalize to many

KKeyon Vafa@keyonV · Jul 11

If you only care about orbits, Newton didn't add much. His laws give the same predictions. But Newton's laws went beyond orbits: the same laws explain pendula, cannonballs, and rockets. This motivates our framework: Predictions apply to one task. World models generalize to many

1.0K

Prof. Nikolai Slavov@slavov_n · Jul 18

𝐌𝐨𝐨𝐫𝐞'𝐬 𝐋𝐚𝐰'𝐬 𝐠𝐫𝐨𝐰𝐭𝐡 𝐫𝐚𝐭𝐞 𝐰𝐚𝐬 𝐧𝐨𝐭 𝐡𝐢𝐠𝐡. Its most remarkable feature was the time span: steady growth over 4 - 5 decades. If comparing to Moore's Law, consider the timescale: 𝐈𝐬 𝐭𝐡𝐞 𝐠𝐫𝐨𝐰𝐭𝐡 𝐬𝐮𝐬𝐭𝐚𝐢𝐧𝐞𝐝 𝐨𝐯𝐞𝐫 𝐝𝐞𝐜𝐚𝐝𝐞𝐬? .

slavov_n's tweet image. 𝐌𝐨𝐨𝐫𝐞'𝐬 𝐋𝐚𝐰'𝐬 𝐠𝐫𝐨𝐰𝐭𝐡 𝐫𝐚𝐭𝐞 𝐰𝐚𝐬 𝐧𝐨𝐭 𝐡𝐢𝐠𝐡.

Its most remarkable feature was the time span: steady growth over 4 - 5 decades.

If comparing to Moore's Law, consider the timescale:

𝐈𝐬 𝐭𝐡𝐞 𝐠𝐫𝐨𝐰𝐭𝐡 𝐬𝐮𝐬𝐭𝐚𝐢𝐧𝐞𝐝 𝐨𝐯𝐞𝐫 𝐝𝐞𝐜𝐚𝐝𝐞𝐬?

.

1.0K

Prof. Nikolai Slavov@slavov_n · Jul 18

Cell size varies by many orders of magnitude. The mechanisms that generate and maintain this extraordinary diversity of sizes remain incompletely understood.

slavov_n's tweet image. Cell size varies by many orders of magnitude.

The mechanisms that generate and maintain this extraordinary diversity of sizes remain incompletely understood.

421

183

27.0K

Prof. Nikolai Slavov@slavov_n · Jul 17

Last week, I talked about technological progress in single-cell proteomics & the biological analysis that it enables: 𝐒𝐢𝐧𝐠𝐥𝐞-𝐜𝐞𝐥𝐥 𝐩𝐫𝐨𝐭𝐞𝐨𝐦𝐢𝐜𝐬: 𝐅𝐫𝐨𝐦 𝐏𝐫𝐨𝐭𝐞𝐢𝐧 𝐕𝐚𝐫𝐢𝐚𝐭𝐢𝐨𝐧 𝐭𝐨 𝐁𝐢𝐨𝐥𝐨𝐠𝐢𝐜𝐚𝐥 𝐅𝐮𝐧𝐜𝐭𝐢𝐨𝐧𝐬 youtu.be/8nhjKSTqeOs?si…

4.0K

Prof. Nikolai Slavov@slavov_n · Jul 16

Increasing throughput is particularly helpful when it also increases data quality. 🚀

PProf. Nikolai Slavov@slavov_n · May 28

I am very enthusiastic for improving fragmentation spectra & the research this empowers. This means increasing the sensitive and scope of sequencing the immunopeptidome ==> better cell therapies. biorxiv.org/content/10.110… 1/2

1.0K

Prof. Nikolai Slavov@slavov_n · Jul 15

The noncanonical inflammasome provides a specialized and direct pathway for host cells to detect intracellular bacterial invasion and initiate a rapid, lytic cell death response to eliminate infected cells.

slavov_n's tweet image. The noncanonical inflammasome provides a specialized and direct pathway for host cells to detect intracellular bacterial invasion and initiate a rapid, lytic cell death response to eliminate infected cells.

2.0K

Prof. Nikolai Slavov@slavov_n · Jul 15

This is my definition of a foundation model.

KKeyon Vafa@keyonV · Jul 11

Newton's laws are a kind of foundation model. They provide a place to start when working on new problems. A good foundation model should do the same. The No Free Lunch Theorem motivates a test: Every foundation model has an inductive bias. This bias reveals its world model.

1.0K