Esther Mathew

Generating Novel Compounds from Mass Spectrometry (MS) Data

Background: MS experiments create mass spectra which show the abundance of different ions with a given mass-to-charge (m/z) ratio.

Goal: Identify substances in a sample from mass spectrum using ML models.

My model uses autoencoder and transformer architectures to generate molecular strings (SMILES) from MS data.

• Tools: PyTorch, DeepSMILES
• Models: Autoencoder, GRU, Transformer, CNN, LSTM

Molecular Graph Tools

Background: Molecular structure analysis is essential in computational chemistry for understanding chemical properties and reactions. This project implements tools to represent molecules as graphs, detect aromatic structures, and enable molecular comparisons.

Goal: Create a computational toolkit that represents molecules as mathematical graphs to detect structural patterns, generate molecular fingerprints, and perform substructure searches.

Key Components

• Graph-based molecular representation with atom nodes and bond edges
• Aromatic ring detection using Hückel's rule (4n+2 π-electrons)
• Molecular fingerprinting algorithm for structure comparison
• Fast substructure search using fingerprint bit matching
• Interactive molecular visualization with CPK coloring

• Tools: Python (NetworkX, NumPy, Matplotlib), C++, Eigen

Leadership YouTube Video

Background: I created a video discussing some helpful leadership concepts that span a variety of topics, including management, communication, and strategy. I learned these topics through a leadership course taught at UC Berkeley by Mike Cheng.

Goal: Create a resource to share this information with a larger community and build on my knowledge throughout my career.