Thanks Nicholas, all credit goes to my students and colleagues, they are the true heroes! 😊

You both make legendary videos :)

Hey thanks for the big support!

Thanks for the kind words! :)

Sawatdee krub, glad to hear that the video was helpful and good luck in your PhD endeavors :)

Thanks, glad you think so, this means a lot :)

@@DataProfessor Thanks, data professor! I recently obtained my PhD degree in theoretical astrophysics, and I'm interested in anti-aging / anti-cancer drug discovery. I don't have a background in biology, while I find your videos super helpful. I used your machine learning scripts from the video "Python for Bioinformatics - Drug Discovery Using Machine Learning and Data Analysis", trying to find compounds that may have potential to inhibit mtor (target CHEMBL2842). The random forest model using 4727 training examples achieved Coefficient of determination (R^2): 0.90 and Mean squared error (MSE): 0.18. In app.py, I had to use -Xms1G -Xmx1G instead of -Xms2G -Xmx2G to generate descriptors_output.csv for just my input molecules, otherwise I'm getting a list of all 4727 compounds in the app. As a test, I run app.py, input CHEMBL413 (Rapamycin) and CHEMBL295124 (Berberine). I got predicted pIC50 = 7.06 and 6.43, respectively. Since these are well known mtor inhibitors, the result looks reasonable. The weird thing happens when I took your original example input in Part VI, example_acetylcholinesterase.txt, which includes 5 candidates AChE inhibitors (CHEMBL133897, CHEMBL336398, CHEMBL131588, CHEMBL130628, CHEMBL130478). And surprisingly I got predicted pIC50 for mtor inhibition in the range 6.5-6.8. I wouldn't expect AChE inhibitors would also inhibit mtor efficiently. Did I do something wrong? It seems that the feature representation is super important in QSAR models. So far we have been using 881 molecular fingerprints. I wonder if we were to use graph neural networks to learn the feature representation of molecules, instead of using hard coded fingerprint features, would the machine learning model achieve better performance?

Awesome! Lots of potential areas to explore in the field.

Thanks!

Thanks krub Boyd, Glad to hear that! You're also doing great things in the data space with your contents 😊

You're very welcome!

I appreciate that!

Glad you enjoyed it!

You’re welcome!

That's a great question, I think it's a great idea, it could be potentially used for representing molecules as interconnected graphs where atoms are nodes and bonds are edges. This could also be applied for biological networks at the systems biology level as well.

Many thanks!

Thank you, glad to hear that 😊

That is awesome! I actually have related hour long lectures: - How to build bioinformatics tools kzfaq.info/get/bejne/gq59Y6ag2tyUp2g.html - Data science for bioinformatics kzfaq.info/get/bejne/aaehhNFqmbG6iok.html - AMA about bioinformaitcs kzfaq.info/get/bejne/eMh_iZZkqJvTeps.html - 6-part series on Bioinformatics from Scratch kzfaq.info/get/bejne/ptKGf7WS2M7FcWw.html

Great idea, will try to simplify as much as possible so that it also appeals to the general data and machine learning community 😊

@@DataProfessor thank you sir. You are an inspiration 👍

Thanks for the support!

Thanks, glad to hear that :)

Hi, the easiest way is to compile a dataset of receptor and ligand pairs along with their respective binding energy. Then apply ML to make the prediction. Instead of binding energy, you could find experimental binding values such as the binding KD or KA (dissociation and association) as well. Additionally, instead of binding, you could predict the inhibition value such as IC50 or pIC50 (the log form). I've made the Bioinformatics video series showing how you can compile your own dataset from the ChEMBL database here kzfaq.info/get/bejne/ptKGf7WS2M7FcWw.html Hope this helps.

@@DataProfessor thank prof. If I have a doubt, I will contact you

Thanks for tuning in!

Hey thanks!

One can use GANs for generating new molecules. If one wants to generate molecular decoys, there’s the DUD-E tool by the Shoichet group. The Reymond group also has done some extensive work in computational enumerating new molecules in a de novo fashion.

@@DataProfessor thanks a ton! I'm gonna look into it.

​@@DataProfessor Thanks for your valuable information, My PhD supervisors suggest to take GANs as my PhD subject and they suggest using GANs for drug discovery ? Is this a good subject for PhD, especially i have low knowledge in cheminformatics and bioinformatics. Would you recommend using GANs to create a synthetic dataset as PhD subject or GANs becomes so an interesting topic for only large companies.

@@ahmedyamany5065 It’s an interesting topic that is at the fore front of the field, both for academia and pharma industry.

I have 2 recommendations, choose 1: Python or R. Please watch the following videos for more info: 1. Python vs R kzfaq.info/get/bejne/sN-YZauGlsvKn2g.html 2. Python 101 playlist kzfaq.info/get/bejne/ir-Aqs6Xzdbbe4k.html 3. R 101 playlist kzfaq.info/get/bejne/rNNxnbqUprG5oYU.html I use both R and Python for my research. The decision is primarily because my students either uses R or Python. Both have cheminformatics related libraries and packages, which are the main reason for using either language. Select a library or package that you think answers your needs. Hope this helps.

@@DataProfessor Thanks, sir. Is it necessary to learn C/C++ in relation to bioinformatics/chemoinformatics study?

Hi, there are several ways of doing this. One of many would be AutoGrow which enumerates molecule in the confinement of a protein binding cavity. There’s also one from KNIME www.knime.com/knime-applications/chemical-library-enumeration

@@DataProfessor thanks for your quick answer. I'm going to review the AutoGrow and KNIME documentation.

You can find me on Twitter and LinkedIn. Details in the video description and also sign up to my newsletter.

Hi, it's actually recorded from a class I taught at the university.

@@DataProfessor your students are so blessed to have you. Anyway, thank you for sharing this. This is a really great help for us who search for resources and consultants online.

Hi, I’ve checked again and the audio works normally

Awesome, I'll soon have some public datasets that the entire community can contribute to. Soon.

@@DataProfessor I would like to do a PhD in Computational Biology and Bioinformatics. Can you mentor me?