Why number of interactions decreased in current string database for plasmodium

Question

[rohitsatyam102]

Hi Everyone

I downloaded the 3 PPI interaction files of Plasmodium from String DB (the current PPI from 36329.protein.links.v12.0-July26_2023 and two more from years 2020 and 2021 i.e. 5833.protein.links.v11.5-Aug12_2021 and 5833.protein.links.v11.0-Oct17_2020) and converted the old Pf IDs to PF3D7 ids. Now when I count the number of interactions of a few genes I am working on I observe less interaction in the current release as compared to the 2020 or 2021 release. Can anyone help me understand why is that the case?

Note: self interaction i.e A-A were removed and only one of symmetrical interaction A-B or B-A was retained before counting the interactions.

[rohitsatyam102]

I understand that the taxon IDs are different and that 5833 represents all Plasmodium falciparum and 36329 represent Pf3D7 but the interaction that previously existed shouldn't just disappear.

[GenoMax]

but the interaction that previously existed shouldn't just disappear.

Perhaps some new information came to light that warranted the correction/removal? You can email String DB folks with specific example and ask.

Answer 1

[damian.szk]

Dear Rohit,

STRING team here. Thanks for the feedback.

Generally, the raw counts of links (edges) per protein may not provide much insight. Most links in STRING have low scores, and relatively minor calibration changes can appear to significantly alter these counts without substantially impacting the network's ability to accurately represent biological processes.

However, the situation with P. falciparum is a bit different.

Each update of STRING involves updating proteomes, sources, and algorithms, by design it’s not meant to only add interactions but improve the network's reflection of the organism's biology. In the latest update, we've completely overhauled the co-expression pipeline (see recent paper), significantly affecting the P. falciparum network.

In STRING, all link sources are traceable. Looking at the past version network of P. falciparum, it's evident that a vast majority of links originated from co-expression predictions. Having some experience with these networks it appears to be the number of links generated by the previous co-expression pipeline is elevated. Here I have greater confidence in the updated network's accuracy. However if the previous version align better with your perspective, you are welcome to use it.

I hope this clarifies your concerns, and thank you again for your feedback.

Answer 2

[Prash]

It's time, the databases honoured Minimum Information About Bioinformatics Investigation ( MiABi) guidelines

STRING, however is dynamic and adheres to huge compendium of databases