Our inspiration is our professor who described the current world scenario of HIV infections and the population diagnosed with AIDs. The constant aim of developing something for social good was a constant source of motivation.
What it does
Our project is an HIV regulator that can in fact detect the HIV drugs from their molecular structure. It performs feature extraction by constructing graph structure representing the molecules and calculating 1D 2D and 3D descriptors for the molecules.
How I built it
We have used special libraries like Pytorch, Tensorflow, Rdkit, Torch, etc to build the model using proper datasets. All references have been provided within the documentation.
Challenges I ran into
We faced various issues like which classification algorithm or which classifier to be used for the prediction and after cross-validating various classifiers, coming up with a particular one was difficult as there hasn't been much work upon this earlier.
Accomplishments that I'm proud of
We were finally proud to have achieved a promising accuracy and better F-scores. We were able to work out the way we wanted the data to be processed and augmented.
What I learned
We had a great time learning about the various molecules and their structure. How we implemented graph structures upon 3D molecules was really a great thing I learned.
What's next for HIV drug prediction model
We will make this better to look upon by using a notebook like Jupyter Notebook or Colabs to give a clean format of all the files implemented together and work upon improvising the algorithm with CNN classifier along with Class Activation Maps.