{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# Shape of Molecules #\n", "\n", "In this notebook we provide an innovative pipeline that makes it possible to find interesting and meaningful structural features for chiemical compounds by exploiting the package $\\href{https://giotto.ai/}{giotto learn}$. The task of this notebook is to classify chemical compunds as HIV inhibitors or non-inhibitors for the HIV virus. The problem is a benchmark for molecules graph representation as stated in this $\\href{https://pubs.rsc.org/en/content/articlehtml/2018/sc/c7sc02664a}{paper}$. The novel idea of this notebook is that of exploiting heat diffusion defined over any-order graph cliques (here nodes and edges) to embed the entire graph. In order to defined such diffusion processes we use the definition of higher-order laplcians matrices of clique complexes (special case of simplicial complexes).\n", "\n", "### Example of heat diffusion over nodes sampled at two different points ###\n", "\n", "