Seminar Series

The Chemical and Biochemical Engineering Department Graduate Student Seminar series is a bi-weekly seminar taking place during the Fall and Winter terms. Each seminar features a different student presenters.

Fall 2025 - Winter 2026: Seminars will be held from 12:30 pm - 1:30 pm every second Thursday. The location will be announced. *Please bring Student Card for attendance*

Speakers to Present in Winter 2026

Feb. 5, 2026, 12:30 pm

Location: ACEB 1415 

Presentation #1: A Multimodal Semi-Supervised Learning Framework for Pharmaceutical Cocrystals Prediction

Presenter: Mohammad Amin Ghanavati

Supervisor(s): Dr. Sohrab Rohani

 Abstract:

Cocrystal formation is a key solid-state engineering approach with broad relevance, especially in pharmaceutical development, where it is widely exploited to modulate solubility, stability, and bioavailability of a substantial fraction of marketed and developing drug molecules that otherwise exhibit poor physicochemical properties. Despite this potential, identifying suitable coformer combinations remains a laborious and uncertain task, and recent data-driven models designed to support discovery are constrained by a fundamental data limitation: experimental databases overwhelmingly emphasize successful cocrystals, while failed attempts (physical mixtures) are sparsely documented. This imbalance biases supervised learning models toward optimistic outcomes and undermines their robustness and reliability.

 To address this challenge, we introduce a two-stage multimodal learning framework aimed at promoting balanced learning while minimizing reliance on costly computations. In the first stage, a multimodal self-supervised pretraining strategy aligns graph-based molecular embeddings from a Graph Attention Network (GAT) with electrostatic potential (ESP) histogram representations, enabling the GAT encoder to learn chemically meaningful features without labelled data. The pretrained encoder is then incorporated into a multimodal ensemble to produce high-confidence pseudo-labels, achieving near-perfect reliability in high-confidence regions (BACC ≈ 0.98). In the second stage, a semi-supervised learning scheme progressively augments reliable pseudo-negative samples, mitigating dataset imbalance and enhancing generalization to chemically distinct systems. This approach improves GAT balanced accuracy from 0.78 to 0.86 on test data, and when evaluated against state-of-the-art deep learning methods on a chemically diverse external benchmark, the resulting GAT-SSL model attains 82.3% balanced accuracy, outperforming prior deep learning methods by up to 20% and exceeding DeepCocrystal by 4.3%. Notably, it delivers the most favorable recall–specificity trade-off (70.5% recall / 94% specificity) while maintaining high precision (86.7%), surpassing DeepCocrystal by 18.7% in precision, all without relying on computationally intensive DFT calculations at inference.

 

Presentation #2: Controlled drug release investigation using metal-organic framework-based prodrug

Presenter: Ehsan Binaeian 

Supervisor(s): Dr. Sohrab Rohani

Abstract:

Chemotherapy and oral administration remain the most widely used modalities in cancer treatment; however, they suffer from serious limitations, including severe side effects, burst release under acidic conditions, premature drug release, and leakage before reaching tumor tissues. To address these challenges, drug delivery systems have attracted significant attention. Prodrugs remain inactive until they are converted into active drugs at specific target sites, thereby minimizing premature release. Metal–organic frameworks (MOFs), as crystalline porous materials, show great potential for drug loading and delivery due to their high porosity, large surface area, tunable structures, functional groups, low toxicity, and good biocompatibility. In this context, MOF-based prodrugs represent promising platforms for controlled anti-cancer drug delivery.

 The main objective of this research is to design and develop zirconium and zinc-based MOF prodrugs by exploiting the functionalization capability of MOFs through tailored modifications, such as the incorporation of disulfide (S–S) and amine (NH₂) groups, followed by methotrexate (MTX) conjugation reactions. Then accumulative release in appropriate media is evaluated properly. The environmental sensitivity of the bonds formed between the MOF topology and MTX is a key factor governing stimuli-responsive drug release. Specifically, upon exposure to stimuli such as the acidic pH of tumor sites or elevated glutathione (GSH) concentrations in cancer cells, approximately ten times higher than in normal cells, drug release is triggered through cleavage of C=N and S–S bonds.