A planned open-source scoring study. Pilot docking and a 0.25 ns solvent-only run are done. 100 ns runs are planned. No lab validation yet.
Hey,
Tuberculosis (TB) is a serious lung infection caused by bacteria. It spreads through the air when a sick person coughs or sneezes, and it still kills over a million people around the world every year. Many patients have to take strong medicines for six months or longer, and some TB bacteria have now become resistant to older drugs like isoniazid. Because of this drug resistance, we need new medicines that attack the bacteria in a direct way.
My name is Navneet Dwivedi. I am an independent researcher from Singrauli, Madhya Pradesh, India. I have a B.Com degree. I also live with SLD, ADHD, Autism, and an auditory learning disability.
I am building TB-Dynamic-Selectivity v3.0. It is a free, open-source tool.
Code:https://github.com/navneetdwivedi147-code/TB-Dynamic-Selectivity
My tool looks for new drug candidates against TB by targeting a bacterial enzyme called InhA (PDB: 4TZK), which the bacteria need to build their outer cell wall. Standard computer docking tests molecules on a frozen picture of this protein. But inside a real human body at 310 K, heat and water make the protein move all the time. My tool checks how long a drug molecule stays near the key Tyr158 spot while the protein moves, and compares target binding with off-target binding in kcal/mol.
This is a computer model only. No laboratory tests have been done yet.
Goal
I want to build a free tool that ranks tuberculosis drug candidates. The tool checks how a molecule behaves against InhA at body temperature (310 K). The goal is to rank candidates better than a standard docking score alone.
How I will do it (if funded)?
If funded, I plan to work on this for three months.
Month 1 (planned): Check my docking outputs. Verify IC₅₀ values from published papers. Rebuild the InhA setups with NAD+. Dock about 50 known TB inhibitors and decoys.
Month 2 (planned): Rent cloud GPUs. Run a speed test. Run 100 ns simulations at 310 K for the top candidates, including MTB-NL4.
Month 3 (planned): Calculate contacts with Tyr158 and NAD+, and buried surface area. Tune the scoring weights. Get a review from an outside structural biologist (not yet identified). Publish the data on Zenodo.
It worked if:
The 100 ns simulations finish without errors.
Reference drugs stay within 3.5 Å of Tyr158 for most frames. Set the exact percentage before the runs begin.
The 310 K score ranks known TB inhibitors better than static Vina docking alone.
An outside structural biologist reviews the setup and confirms that it is correct.
It did not work if:
The simulations cannot finish within the planned budget.
The 310 K score does not rank the inhibitors better than static docking.
The target is $5,000, and the minimum is $2,500. The planned research budget is:
1.Living costs and moving to Lucknow (3 months) – $1,950 ($650/month for rent, food, and bills so I can shift to Lucknow next month and work full-time, six days a week).
Cloud GPU rent (estimated): $1,200, about 2,100 to 2,400 hours at $0.50 to $0.55 per hour (to be confirmed with current RunPod or Vast.ai pricing).
Computer upgrade and backup storage (estimated): $450, for extra RAM, an SSD, and a backup hard drive.
Research software (estimated): $350, for three months of cloud and analysis tools.
Outside structural biology review (conditional): $650. This will be spent only after a qualified reviewer is confirmed.
Bank transfer and currency conversion (estimated): $400.
If only $2,500 is raised, I will reduce the number of complexes from 10 to 12 down to 4 to 6, and spend the money on GPU hours, storage, outside review, and transfer fees.
Month 1 (planned): Check my existing docking outputs (AutoDock Vina 1.2.7, NAD-containing receptor). Make small README clarifications. Verify IC₅₀ values against the published papers. Rebuild holo-InhA setups with NAD+ from the docked pose. Dock about 50 TB inhibitors and decoys (decoys to be selected).
Month 2 (planned): Rent cloud GPUs (RunPod or Vast.ai). Run a speed test on a holo-InhA system similar in size to my test system (about 137,000 atoms). Run 100 ns GROMACS simulations at 310 K for the top candidates, including MTB-NL4 (the number depends on the speed test).
Month 3 (planned): Calculate energy, Tyr158 and NAD+ contacts, and buried surface area. Tune the v3.0 weights (currently heuristic) with grouped cross-validation. Get a review from an outside structural biologist (not yet identified). Upload all data to Zenodo.
The target is $5,000, and the minimum is $2,500. The planned research budget is:
Cloud GPU rent (estimated): $1,200, about 2,100 to 2,400 hours at $0.50 to $0.55 per hour (to be confirmed with current RunPod or Vast.ai pricing).
Computer upgrade and backup storage (estimated): $450, for extra RAM, an SSD, and a backup hard drive.
Research software (estimated): $350, for three months of cloud and analysis tools.
Outside structural biology review (conditional): $650. This will be spent only after a qualified reviewer is confirmed.
Bank transfer and currency conversion (estimated): $400.
The planned research budget totals $3,050. The remaining $1,950 is not yet assigned, and it would go to additional GPU hours.
If only $2,500 is raised, I will reduce the number of complexes from 10 to 12 down to 4 to 6, and spend the money on GPU hours, storage, outside review, and transfer fees.
GPU runs are slower than expected. In that case I will run fewer complexes (4 to 6 instead of 10 to 12), and share the exact timings.
The 310 K motion does not improve ranking over static Vina docking. In that case I will still publish the full dataset and the negative results on GitHub and Zenodo.
I cannot find an outside structural biology reviewer within three months. In that case I will not spend the $650 review budget.
No funding has been received so far. I have planned applications to Emergent Ventures and Experiment.com, and I will update this section once their status is confirmed.
Comments
No comments yet. Sign in to create one!