4/n --> Consistent Frequency
For a general term like "in this study", the results for 2023-2024 were exactly on the trendline based on the 2014-2022 data. So probably, there is no reason to believe that publication numbers rose in general after 2022, but only certain word frequencies changed.

3/n --> Decreased Frequency
Unexpectedly, there was a drop in abstracts using widespread words like "our results" or "we show" after 2022. This might indicate the use of LLMs by researchers, as LLM-generated text often uses passive rather than active language ("this shows" versus "we show").

2/n --> Increased Frequency
Some words are clearly more frequently found in biomedical literature after 2022, such as "pertinent", "treasure trove" or "shedding light". This could be due to overall publication numbers increasing for LLM-unrelated reasons.

I explored whether scientific literature provides evidence of AI use in research publications. I analyzed the use of certain words before & after 2022 using PubMed, focusing on terms commonly used by ChatGPT (e.g., "delve") & standard scientific phrases ("in this study", "our results"). 1/n

Pubmed
#Science

This quote in the article got me thinking a lot. What if each reductionist finding itself is "correct" but that putting it all back together won't work because each reductionist finding came with specific assumptions which will sum up to a lot of noise in the final model.

"Unless you experience the unpleasant symptoms of being wrong, your brain will never revise its models. Before your neurons can succeed, they must repeatedly fail. There are no shortcuts for this painstaking process."

Great to be in San Francisco representing Ksilink during the JP Morgan Healthcare week. Looking forward to lots of learning experiences. If you have questions about neuronal drug discovery models and AI for data analysis, just reach out! #JPM #JPM25

More on post-transcriptional & neuroinflammation regulation in Parkinson's disease. aSyn aggregates trigger inflammatory responses and neuronal toxicity through dysregulated ADAR1-mediated RNA editing in astrocytes. In vitro & post-mortem human PD brain evidence. www.biorxiv.org/content/10.1...

Do the quiz, it's fun:

9/n
Our study therefore shows that ARPC2 inhibition can rescue early PMDS-related hyperdifferentiation phenotypes. We believe that human stem cell models will significantly advance the discovery and evaluation of new, safe neurodevelopmental treatments in the future.

8/n
SHANK3 is a crucial synaptic scaffolding protein, but on the electrophysiological level we did not find any rescuing effects of ARPC2 inhibition. This might be caused by the fact that the SHANK3 protein level inside of synapses remains unchanged after ARPC2 inhibition.

7/n
The ARPC2-targetting compound Benproperine also rescued increased synapse numbers when administered early during differentiation, but had little impact when added later during neuronal differentiation.

6/n
We found that ARPC2 was expressed in out cells and that its inhibition rescued the decreased proliferation in SHANK3-deficient young neurons and rescued ARPC2 colocalization with ß-actin.

4/n
Next we performed a drug screen with >7,000 small molecules with known mode-of-action to rescue the hyperdifferentiation phenotype in CRISPR-engineered SHANK3-deficient cells.

3/n
SHANK3 haploinsufficiency does not only impact early differentiation, but leads to increased MAP2 network length and pre-synaptic spots, variable synapse numbers, and decreased total SHANK3 content and neuronal activity in SHANK3-deficient neurons.

2/n
PMDS arises from mutations in the terminal region of chromosome 22q13, impacting the SHANK3 gene. We found that PMDS patient-derived and CRISPR-engineered cell lines lacking one SHANK3 copy show decreased proliferation & increased differentiation, mirroring traits of PMDS.

In our latest study together with IStem (Paris) we focused on Phelan-McDermid syndrome (PMDS), a genetic disorder that presents complex challenges in neurodevelopment and synaptic function.
#PMDS #StemCells
Read the paper (rdcu.be/dKnYt) & the thread. 1/n