What is Tamarind?
Tamarind is a website that makes computational biology tools available online, for example, protein design, docking, and structure prediction software. As a SAAS (Software as a Service) it streamlines the entire process of downloading, installing, learning, and using these softwares. The founders’ goal with Tamarind is to make drug discovery more accessible and easier to use, allowing researchers to focus on the results and impacts of their work.
By creating an account, you have access to over 50 different softwares. You are able to submit 10 jobs per month. They offer many popular drug development programs, including AutoDock vina, UniFold, and ADMET prediction. Specifically for small molecule docking, they have programs with either a physics or machine learning based approach.
Tamarind home page ->
My Experience with Drug Development
In my most recent research project, I created novel derivatives of the chemotherapy agent Daunorubicin. My goal was to reduce its fatal cardiotoxicity, ultimately reducing the incidence of heart failure. To develop the derivatives, I needed to understand the mechanism of action of the original drug and the location of the binding pocket. I used AutoDock vina and PyMOL to determine the binding pocket and specific hydrogen bonds. I used AutoDock vina to test each of my derivatives’ binding to both proteins in the study, and mainly analyzed the binding score and RMSD. I used AutoDock vina through the visual interface Chimera, as I preferred to work on a UI instead of command prompt.
Learning and mastering these different technologies was a major challenge. Simply downloading and installing the software took days. From finding the software compatible with my computer to parsing through GitHub logs and even consulting with my dad, it took a lot of insider knowledge and time. I spent hours watching different YouTube tutorials, all of which told me slightly different techniques for preparing and binding ligands. From there, I learned how to format all of the input files, model them on the interface, and finally interpret the results. Not to mention, every two hours, my computer would crash!
Another limitation of using AutoDock vina on Chimera was that it was hard to visualize the position of the ligand in the binding pocket after docking. Therefore, I would transfer the complex over to another visualization software called PyMOL, where I could see the amino acids, hydrogen bonds, and strength of the hydrogen bonds involved with the complex.
Testing Tamarind
To test out the programs that Tamarind offers, I decided to replicate the exact process I did from my previous research project, as that is what I have the most knowledge and experience with. I used their AutoDock vina and submitted two jobs- Daunorubicin docking with proteins A and B.
First, I prepared my proteins in Chimera by removing any extra hydrogen molecules and nucleic acids. I also changed my ligands into pdb files through Chimera. Then, I imported the ligand and protein files and modified the exhaustiveness and binding pocket location. Finally, I submitted both jobs on May 25th!
Overall, the experience was very streamlined, easy to understand, and far quicker than using the actual software. I got my first result back on May 26th and the second on June 6th. The output zip file included a settings file, the output scores and RMSD in an Excel spreadsheet, a text file log of the job, and pdb and pdbqt files of the docking positions.
Nevertheless, there were a few limitations of the experience…
- Binding modes
Firstly, it only provides you with 6 binding modes, whereas the actual software provides you up to 10.
- Binding pocket interface
Another limitation was that the interface to select the binding pocket was not ideal. In the interest of making jobs on the actual software and Tamarind as similar as possible, I kept track of the exact coordinates of the binding pocket. However, the interface only allowed me to select a pocket by physically moving the box, so they were not exactly the same. It also didn’t allow me to identify the different chains in the dimer protein, which is crucial to determining the binding pocket.
Although these are simple UI fixes, they would be greatly beneficial to improving the accuracy of jobs.
Protein A
Local AutoDock vina
Tamarind
Protein B
Interestingly, the binding scores from Tamarind were all higher than the scores from my local AutoDock software. I am curious as to whether these discrepancies are due to the slight changes in the binding pocket, or any extra molecule prep that Tamarind does that I did not do manually.
Overall
Overall, I really enjoyed using Tamarind. The workflow is much simpler and easy to understand and cut down the time it took to submit my jobs. Making complex softwares like AutoDock available online makes computational drug development much more accessible to people all around the globe. I am excited to see where Tamarind heads, and I definitely plan to continue using it!
