From Nanoemulsion-Templated Drug Nanoparticles to Machine Learning Property Prediction: The Research Path of Lucas Attia

Getting poorly soluble drugs into the bloodstream through the gut remains one of the most stubborn problems in pharmaceutical development. A recent editorial published in AAPS Open on 10 September 2026 spotlights Lucas Attia, a chemical engineer whose doctoral research at MIT tackled this challenge from the formulation side and who has since moved into machine learning for drug discovery. The editorial, written by Attia himself, traces a research arc that runs from engineering drug nanoparticles in hydrogels to building predictive models of molecular properties, and it offers a detailed look at how a modern formulation scientist moves between experiment, simulation, and data-driven modeling.

The problem Attia set out to address is well documented. Many small-molecule drug candidates are lipophilic, meaning they dissolve poorly in water, and that limited oral bioavailability remains a crucial challenge in drug development. As candidate molecules grow in size and molecular complexity, enabling bioavailability increasingly depends on engineering formulations rather than on redesigning the drug itself. One of the most successful strategies is nanosizing, which has unlocked the oral delivery of drugs such as sirolimus, sold as Rapamune, and aprepitant, sold as Emend.

Yet producing drug nanoformulations is itself difficult. The dominant approach relies on resource-intensive unit operations such as nano-milling, which requires long residence times, exhibits low energy efficiency of less than two percent, and can destabilize the very drug nanoparticles it is meant to produce. In contrast, so-called bottom-up methods directly assemble and stabilize drug nanoparticles, leveraging the benefits of excipients during particle solidification. Attia’s PhD research focused on one such bottom-up process: nanoemulsion templating. In this process, a drug-loaded nanoemulsion is dispersed in a gelling polymer solution, gelled to immobilize the nanodroplets in a hydrogel mesh, and then dried to simultaneously produce and stabilize solid drug nanoparticles in a polymer matrix.

The foundations of the technique had been laid by earlier work in the group of his advisor, Patrick S. Doyle at MIT. Prior studies established the mechanism of droplet isolation and confined crystallization inside the hydrogel and demonstrated the applicability of new gelling materials. However, key structure-property-processing relationships remained underexplored, and the applicability of the process had been demonstrated only for two model hydrophobic actives, fenofibrate and ibuprofen, using a single solvent, anisole. Attia’s thesis addressed these gaps through two complementary research themes: one exploring the structure-property-processing relationships that govern the process, and the other generalizing nanoemulsion templating to new chemistry.

The first theme asked fundamental questions about control over the templated particles. Experimentally, the team observed that the degree of fenofibrate nanoparticle crystallinity depended on the surfactant concentration in the nanoemulsion precursor. Molecular dynamics simulations revealed why: surfactant screened the nanocrystal surface from destabilizing polymer interactions. In a striking methodological finding, the classical molecular dynamics simulations were too short to record crystallization events directly, but short time-scale crystal fluctuations on the order of nanoseconds proved predictive of the experimental crystallinity. That result suggests that brief simulated fluctuations can serve as a computable proxy for a process outcome that would otherwise require far longer simulations or full experiments to observe.

The same theme also sought material levers to control how fast the drug is released from the particles. Attia and colleagues adapted nanoemulsion templating into a core-shell hydrogel particulate system that combined a thermally gelling core with an ionotropically gelling shell. These all-aqueous core-shell hydrogels enabled diverse controlled release profiles, unlocking programmable delayed, sustained, and sequential release of distinct active ingredients.

The second theme pushed the process beyond the model compounds and solvent used previously. In a case study using cannabidiol, clotrimazole, and cholecalciferol, the team demonstrated precise size control between 100 and 1000 nanometers for these chemically diverse lipophilic active ingredients. Importantly, the work reported a rare quantitative validation of the Ostwald-Freundlich relation, the classical theory describing how solubility depends on particle size, finding that the solubility enhancement of cannabidiol nanoparticles is well explained by the theory. Beyond its relevance to formulation, the editorial notes that this validation sheds light on nanoscale thermodynamics more broadly.

A practical bottleneck emerged during this generalization effort. When adapting the process to a new active ingredient, identifying a suitable organic solvent carrier turned out to be the rate-limiting step. In response, Attia developed FASTSOLV, a machine learning model designed to predict organic solubility and accelerate solvent selection. Benchmarked against public solubility datasets, FASTSOLV achieved state-of-the-art performance, and according to the editorial it has since become a widely used tool for in silico solubility screening and selection. Taken together, the editorial argues, the thesis helped mature nanoemulsion templating as a platform technology for formulating drug nanoparticles, with the molecular dynamics work providing general insight into how excipients control nanoparticle structure.

Attia’s interests have since converged toward data-driven modeling of chemical and material properties. Working with Lawrence Berkeley National Laboratory, he contributed to deep learning models that predict optical nanoparticle properties. Through an industrial collaboration, he helped develop a counterfactual-based machine learning framework that conditions model predictions on prior observations at inference time, enabling more accurate predictions within a local chemical or material space. These projects reflect the breadth of expertise he accumulated during his PhD, spanning nanoformulation, soft materials, molecular simulation, and machine learning.

Today, Attia is a machine learning scientist at Eli Lilly and Company, where he builds property prediction models through the collaborative TuneLab platform, deploying models for partner biotech companies to use in their drug discovery programs with a focus on physicochemical and ADME properties. His long-term goal is to develop cheminformatics tools and machine learning representations explicitly designed for formulation science. As he notes in the editorial, formulation scientists recognize that complex formulations are not simple mixtures of their components, yet existing machine learning approaches often represent them as if they were, and capturing the rich physics that governs formulations will be crucial to enabling effective data-driven formulation design. His career has been recognized with numerous awards, including the ACS Colloids and Surface Science Symposium Langmuir Graduate Award in 2024, the AIChE PD2M Student Award in 2025, and the ACS CAS Future Leaders Award in 2026, alongside a PhD in chemical engineering from MIT funded by the Department of Energy Computational Science Graduate Fellowship from 2021 to 2025.

Subject of Research: Nanoemulsion templating for drug nanoparticle formulation and machine learning molecular property prediction

Article Title: Spotlighting rising researcher Lucas Attia: from drug nanoparticle engineering to molecular property prediction

Article References: Attia, L. (2026). Spotlighting rising researcher Lucas Attia: from drug nanoparticle engineering to molecular property prediction. AAPS Open, 12(1), Article 63. https://doi.org/10.1186/s41120-026-00200-3

Image Credits: AI Generated

DOI: 10.1186/s41120-026-00200-3

Keywords: nanoemulsion templating, drug nanoparticles, oral bioavailability, nanosizing, hydrogels, molecular dynamics simulations, Ostwald-Freundlich relation, machine learning, solubility prediction, FASTSOLV, formulation science, Eli Lilly