# Predict molecular properties ## Context This dataset contains molecular properties scraped from the [PubChem database](http://pubchem.ncbi.nlm.nih.gov). Each file contains properties for thousands of molecules , made up of the elements H, C, N, O, F, Si, P, S, Cl, Br, and I. The dataset is related to a [previous one](https://www.kaggle.com/burakhmmtgl/energy-molecule) which had fewer number of molecules, where the features were preconstructed. Instead, this dataset is a challenging case for feature engineering and is subject of active research (see references below). ## Data Description The utilities used to download and process the data can be accessed from my Github [repo](https://github.com/bhimmetoglu/RoboBohr/tree/master/utils). Each JSON file contains a list of molecular data. An example molecule is given below: { 'En': 37.801, 'atoms': [ {'type': 'O', 'xyz': [0.3387, 0.9262, 0.46]}, {'type': 'O', 'xyz': [3.4786, -1.7069, -0.3119]}, {'type': 'O', 'xyz': [1.8428, -1.4073, 1.2523]}, {'type': 'O', 'xyz': [0.4166, 2.5213, -1.2091]}, {'type': 'N', 'xyz': [-2.2359, -0.7251, 0.027]}, {'type': 'C', 'xyz': [-0.7783, -1.1579, 0.0914]}, {'type': 'C', 'xyz': [0.1368, -0.0961, -0.5161]}, {'type': 'C', 'xyz': [-3.1119, -1.7972, 0.659]}, {'type': 'C', 'xyz': [-2.4103, 0.5837, 0.784]}, {'type': 'C', 'xyz': [-2.6433, -0.5289, -1.426]}, {'type': 'C', 'xyz': [1.4879, -0.6438, -0.9795]}, {'type': 'C', 'xyz': [2.3478, -1.3163, 0.1002]}, {'type': 'C', 'xyz': [0.4627, 2.1935, -0.0312]}, {'type': 'C', 'xyz': [0.6678, 3.1549, 1.1001]}, {'type': 'H', 'xyz': [-0.7073, -2.1051, -0.4563]}, {'type': 'H', 'xyz': [-0.5669, -1.3392, 1.1503]}, {'type': 'H', 'xyz': [-0.3089, 0.3239, -1.4193]}, {'type': 'H', 'xyz': [-2.9705, -2.7295, 0.1044]}, {'type': 'H', 'xyz': [-2.8083, -1.921, 1.7028]}, {'type': 'H', 'xyz': [-4.1563, -1.4762, 0.6031]}, {'type': 'H', 'xyz': [-2.0398, 1.417, 0.1863]}, {'type': 'H', 'xyz': [-3.4837, 0.7378, 0.9384]}, {'type': 'H', 'xyz': [-1.9129, 0.5071, 1.7551]}, {'type': 'H', 'xyz': [-2.245, 0.4089, -1.819]}, {'type': 'H', 'xyz': [-2.3, -1.3879, -2.01]}, {'type': 'H', 'xyz': [-3.7365, -0.4723, -1.463]}, {'type': 'H', 'xyz': [1.3299, -1.3744, -1.7823]}, {'type': 'H', 'xyz': [2.09, 0.1756, -1.3923]}, {'type': 'H', 'xyz': [-0.1953, 3.128, 1.7699]}, {'type': 'H', 'xyz': [0.7681, 4.1684, 0.7012]}, {'type': 'H', 'xyz': [1.5832, 2.901, 1.6404]} ], 'id': 1, 'shapeM': [259.66, 4.28, 3.04, 1.21, 1.75, 2.55, 0.16, -3.13, -0.22, -2.18, -0.56, 0.21, 0.17, 0.09] } 1. **En**: This field is the molecular energy calculated using a force-field method. See references [1,2] for details. This is the target variable which is being predicted. 2. **atoms**: This field contains the name of the element and the position (x,y,z coordinates) and needs to be used for feature engineering. 3. **id** : This field is the PubChem Id 4. **shapeM** : This field contains the shape multipoles and can be used for feature engineering. For definition of shape multipoles, see reference [3]. Notice that each molecule contains different number and types of atoms, so it is challenging to come up with features that can describe every molecule in a unique way. There are several approaches taken in the literature (see the references), one of which is to use the Coulomb Matrix for a given molecule defined by $$ C_{IJ} = rac{Z_I Z_J}{\vert R_I - R_J \vert}, \quad ({\rm I eq J}) \qquad C_{IJ} = Z_I^{2.4}, \quad (I=J) $$ where $Z_I$ are atomic numbers (can be looked up from the periodic table for each element), and ${\vert R_I - R_J \vert}$ is the distance between two atoms I and J. The previous [dataset](https://www.kaggle.com/burakhmmtgl/energy-molecule) used these features for a subset of molecules given here, where the maximum number of elements in a given molecules was limited by 50. There are around 100,000,000 molecules in the whole database. As more files are scraped, new data will be added in time. Note: In the previous [dataset](https://www.kaggle.com/burakhmmtgl/energy-molecule), the molecular energies were computed by quantum mechanical simulations. Here, the given energies are computed using another method, so their values are different. ## Inspiration Simulations of molecular properties are computationally expensive. The purpose of this project is to use machine learning methods to come up with a model that can predict molecular properties from a database. In the PubChem database, there are around 100,000,000 molecules. It could take years to do simulations on all of these molecules, however machine learning can be used to predict their properties much faster. As a result, this could open up many possibilities in computational design and discovery of molecules, compounds and new drugs. This is a regression problem so mean squared error is minimized during training. I am looking for Kagglers to find the best model and reduce mean squared error as much as possible! ## References [1] Halgren TA. Merck Molecular Force Field: I. Basis, Form, Scope, Parameterization and Performance of MMFF94. J. Comp. Chem. 1996;17:490-519. [2] Halgren TA. Merck Molecular Force Field: VI. MMFF94s Option for Energy Minimization Studies. J. Comp. Chem. 1999;20:720-729. [3] Kim, Sunghwan, Evan E Bolton, and Stephen H Bryant. “PubChem3D: Shape Compatibility Filtering Using Molecular Shape Quadrupoles.” J. Cheminf. 3 (2011): 25. [4] Himmetoglu B.: Tree based machine learning framework for predicting ground state energies of molecules, J. Chem. Phys 145, 134101 (2016) [5] Rupp M., Ramakrishnan R., von Lilienfeld OA.: Machine Learning for Quantum Mechanical Properties of Atoms in Molecules, J. Phys. Chem. Lett. , 6(16): 3309–3313 (2015) [6] Montavon G., Rupp M., Gobre V., Vazquez-Mayagoitia A., Hansen K., Tkatchenko A., Müller K-R., von Lilienfeld OA.: Machine learning of molecular electronic properties in chemical compound space, New J. Phys., 15(9): 095003 (2013) [7] Hansen K., Montavon G., Biegler F., Fazli S., Rupp M., Scheffler M., von Lilienfeld OA., Tkatchenko A., Müller K-P.: Assessment and Validation of Machine Learning Methods for Predicting Molecular Atomization Energies, J. Chem. Theory Comput. , 9(8): 3543–3556 (2013)