Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session D

Poster A2 - Machine Learning Supported Identification of Hydrophobic Drugs to Use in Long-Acting Injectable Formulations

Friday, October 25, 2024

3:30 PM - 4:30 PM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Kent Qi (he/him/his)

Visiting Student Researcher
Brigham and Women's Hospital, Harvard Medical School
Mississauga, Ontario, Canada

Co-Author(s)

FC

Flavia Codreanu

Visiting Student Researcher
Brigham and Women's Hospital, Harvard Medical School
Mississauga, Ontario, Canada
BC

Benjamin Clark

Research Engineer
Brigham and Women's Hospital, Harvard Medical School, Massachusetts Institute of Technology, United States
CC

Colin Cotter

Research Engineer
Brigham and Women's Hospital, Harvard Medical School, Massachusetts Institute of Technology, United States
SL

Susan R. Ling

Visiting Student Researcher
Brigham and Women's Hospital, Harvard Medical School, United States
JL

Jia Liang

Technical Associate
Brigham and Women's Hospital, Harvard Medical School, Massachusetts Institute of Technology, United States
PK

Parmiss Khosravi

Technical Associate
Brigham and Women's Hospital, Harvard Medical School, Massachusetts Institute of Technology, United States
AL

Aaron Lopes

Senior Research Engineer
Brigham and Women's Hospital, Harvard Medical School, Massachusetts Institute of Technology, United States
Yogesh Deshpande, MS Mechanical Engineering (he/him/his)

Senior Research Engineer
Brigham and Women's Hospital, Harvard Medical School, United States
Nabil Shalabi, PhD

Instructor in Medicine
Brigham and Women's Hospital, Harvard Medical School, Massachusetts Institute of Technology, United States

Last Author(s)

JL

Jason Li (he/him/his)

Instructor in Medicine
Brigham and Women's Hospital, Harvard Medical School, Massachusetts Institute of Technology
Cambridge, Massachusetts, United States

Primary Investigator(s)

GT

Giovanni Traverso

Associate Professor
Brigham and Women's Hospital, Harvard Medical School, Massachusetts Institute of Technology, United States

Introduction: Chronic health conditions often necessitate ongoing drug therapies, which typically require frequent oral or injectable administrations. However, adherence rates to these long-term therapies are alarmingly low, estimated at around 50% even in developed nations. Reducing the frequency of dosing events has been shown to significantly improve adherence, prompting a shift towards the development of extended-release drug delivery systems.

Injectable microcrystalline suspension drug depots, such as Depo-Provera, Sayanna Press, and Apretude, exemplify this approach. These long-acting injectables (LAIs) utilize microcrystalline drug formulations suspended in a liquid medium, where the drug crystals gradually dissolve, releasing the active pharmaceutical ingredient (API) over time. The therapeutic duration of these formulations is directly proportional to the initial crystal size. Typically, these LAIs are derived from Biopharmaceutical Classification System (BCS) class II and IV drugs, which are characterized by low water solubility. Due to their extended-release nature and limitations in drug loading capacity, the APIs used must be both stable and potent.

Our research leverages machine learning to screen and identify suitable API candidates based on their melting points and hydrophobicity. We then conduct in-vitro tests on a subset of these candidates to validate their potential for use in LAI formulations. Through this rigorous process, we identified and validated two promising API candidates, Brexpiprazole and Pimozide, for development into LAI antipsychotics. Given that hydrophobic drugs constitute a significant portion of small molecule therapeutics, our approach holds substantial promise for enhancing patient adherence and improving treatment outcomes.

Materials and

Methods: A comprehensive library of all FDA-approved small molecule API was compiled from the DrugBank database [1]. Each entry was represented by its chemical name, its simplified molecular-input line-entry system (SMILES) fingerprint, molecular weight, logP, and, where available, experimental values for melting temperature and water solubility. Duplicate entries, based on the SMILES fingerprint, were removed from the library. The OpenSol XGBoost model machine learning algorithm was subsequently employed to predict the melting temperature and water solubility for each molecule [2]. Each molecule was subsequently identified as a hit if the following criteria was satisfied: Melting temperature > 37ºC, solubility in water ≤10 µg/ml.
Identified hits were organized by indication. Any API matching our indication of interest was selected for validation. Levonorgestrel (LNG), an API commonly used in contraceptive LAIs, was tested along the candidate APIs and served as a positive control.

In-vitro solubility of APIs was assessed by adding excess powder of each molecule to a pH 7.4 phosphate buffer solution (PBS). Samples were then incubated at 37°C under constant agitation for 48 hours. In-vitro stability of APIs was assessed by incubating excess powder of each molecule to a pH 7.4 PBS. Samples were then incubated at 37°C and 60°C without agitation. Samples were collected over a 3-month period. In-vitro release kinetic of APIs was evaluated by adding drug in accelerated release media composed of 10% sodium dodecyl sulfate in pH 7.4 PBS. Samples were the 37°C under constant agitation, collected over a 1-month period. All sample were analyzed using RP-HPLC.

Results, Conclusions, and Discussions: The screening process began with a dataset of 2579 small molecule compounds obtained from the DrugBank database. The first exclusion criterion removed compounds with melting points below 37ºC. The melting point is screened for as the drug must remain crystalline under physiological temperatures. This criterion excluded 110 compounds, leaving 2469 candidate compounds. The second exclusion criterion removed compounds with water solubility greater than 10 µg/ml. The focus was on identifying highly insoluble compounds. Although the screen could potentially include slightly insoluble or soluble compounds, this specific threshold was set to ensure the selection of compounds with minimal solubility. This step excluded 1917 compounds, resulting in 552 remaining candidates. The final exclusion criterion was to select compounds relevant to an indication of interest. Our focus was on compounds indicated for chronic mental health conditions. Applying this criterion narrowed down the selection to three final candidates acquired for further screening: Brexpiprazole, Flupentixol, and Pimozide.

In-vitro solubility studies indicated that Brexpiprazole, Pimozide, and LNG were below the quantifiable limits of detection of RP-HPLC, aligning with the expected low water solubility from the in-silico screen. However, Flupentixol exhibited significantly higher solubility than predicted, highlighting the necessity for in-vitro validation. Consequently, only Brexpiprazole and Pimozide were further evaluated. In-vitro stability studies demonstrated that both Brexpiprazole and Pimozide exhibited minimal degradation. In-vitro release studies in accelerated media revealed that both drugs exhibited sustained release kinetics.

The screening process successfully identified Brexpiprazole and Pimozide as promising candidates for LAI formulations for chronic mental health conditions. Both Brexpiprazole and Pimozide demonstrated favorable stability and release kinetics, supporting their potential for formulation into a LAI.

Acknowledgements (Optional):

References:
[1] C. Knox et al., “DrugBank 6.0: the DrugBank Knowledgebase for 2024,” Nucleic acids research, vol. 52, no. D1, pp. D1265–D1275, 2024, doi: 10.1093/nar/gkad976.

[2] X. Zhu et al., “Building Machine Learning Small Molecule Melting Points and Solubility Models Using CCDC Melting Points Dataset,” Journal of chemical information and modeling, vol. 63, no. 10, pp. 2948–2959, 2023, doi: 10.1021/acs.jcim.3c00308.