Rıza Özçelik
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Rıza Özçelik has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
- Francesca Grisoni
Eindhoven University of Technology P.O. Box 513
13 shared publications - Arzucan Özgür
Department of Computer Engineering, Boğaziçi University, İstanbul, Turkey
4 shared publications - Emanuele Criscuolo
Eindhoven University of Technology
4 shared publications Emanuele Criscuolo
Department of Biomedical Engineering, Institute for Complex Molecular System (ICMS) and Eindhoven AI Systems Institute (EAISI)
3 shared publications- Abdullatif Köksal
Department of Computer Engineering, Boğaziçi University, İstanbul, Turkey
2 shared publications - Christian Ottmann
Eindhoven University of Technology
2 shared publications
Latest Publications
- Open Access
- 8 August 2025
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
Identifying 14-3-3 interactome binding sites with deep learning
Protein–protein interactions are at the heart of biological processes. Understanding how proteins interact is key for deciphering their roles in health and disease, and for therapeutic interventions. However, identifying protein interaction sites, especially for intrinsically disordered proteins, is challenging. Here, we developed a deep learning framework to predict potential protein binding sites to 14-3-3 – a ‘central hub’ protein holding a key role in cellular signaling networks. After systematically testing multiple deep learning approaches to predict sequence binding to 14-3-3, we developed an ensemble model that achieved a 75% balanced accuracy on external sequences. Our approach was applied prospectively to identify putative binding sites across medically relevant proteins (ranging from highly structured to intrinsically disordered) for a total of approximately 300 sequences. The top eight predicted peptide sequences were experimentally validated in the wet-lab, and binding to 14-3-3 was confirmed for five out of eight sequences (Kd ranging from 1.6 ± 0.1 μM to 70 ± 5 μM). The relevance of our results was further confirmed by X-ray crystallography and molecular dynamics simulations. These sequences represent potential new binding sites within the 14-3-3 interactome (e.g., relating to Alzheimer's disease as the binding to tau is not the new part), and provide opportunities to investigate their functional relevance. Our results highlight the ability of deep learning to capture intricate patterns underlying protein–protein interactions, even for challenging cases like intrinsically disordered proteins. To further the understanding and targeting of 14-3-3/protein interactions, our model was provided as a freely accessible web resource at the following URL: https://14-3-3-bindsite.streamlit.app/.
- 8 August 2025
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
Open AccessIdentifying 14-3-3 interactome binding sites with deep learning
Protein–protein interactions are at the heart of biological processes. Understanding how proteins interact is key for deciphering their roles in health and disease, and for therapeutic interventions. However, identifying protein interaction sites, especially for intrinsically disordered proteins, is challenging. Here, we developed a deep learning framework to predict potential protein binding sites to 14-3-3 – a ‘central hub’ protein holding a key role in cellular signaling networks. After systematically testing multiple deep learning approaches to predict sequence binding to 14-3-3, we developed an ensemble model that achieved a 75% balanced accuracy on external sequences. Our approach was applied prospectively to identify putative binding sites across medically relevant proteins (ranging from highly structured to intrinsically disordered) for a total of approximately 300 sequences. The top eight predicted peptide sequences were experimentally validated in the wet-lab, and binding to 14-3-3 was confirmed for five out of eight sequences (Kd ranging from 1.6 ± 0.1 μM to 70 ± 5 μM). The relevance of our results was further confirmed by X-ray crystallography and molecular dynamics simulations. These sequences represent potential new binding sites within the 14-3-3 interactome (e.g., relating to Alzheimer's disease as the binding to tau is not the new part), and provide opportunities to investigate their functional relevance. Our results highlight the ability of deep learning to capture intricate patterns underlying protein–protein interactions, even for challenging cases like intrinsically disordered proteins. To further the understanding and targeting of 14-3-3/protein interactions, our model was provided as a freely accessible web resource at the following URL: https://14-3-3-bindsite.streamlit.app/.
- Top Cited PapersOpen Access
- 9 July 2025
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 65 (14) , 7352-7372
Generative Deep Learning for de Novo Drug Design─A Chemical Space Odyssey
In recent years, generative deep learning has emerged as a transformative approach in drug design, promising to explore the vast chemical space and generate novel molecules with desired biological properties. This perspective examines the challenges and opportunities of applying generative models to drug discovery, focusing on the intricate tasks related to small molecule generation, evaluation, and prioritization. Central to this process is navigating conflicting information from diverse sources─balancing chemical diversity, synthesizability, and bioactivity. We discuss the current state of generative methods, their optimization, and the critical need for robust evaluation protocols. By mapping this evolving landscape, we outline key building blocks, inherent dilemmas, and future directions in the journey to fully harness generative deep learning in the "chemical odyssey" of drug design.
- 9 July 2025
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 65 (14) , 7352-7372
Top Cited PapersOpen AccessGenerative Deep Learning for de Novo Drug Design─A Chemical Space Odyssey
In recent years, generative deep learning has emerged as a transformative approach in drug design, promising to explore the vast chemical space and generate novel molecules with desired biological properties. This perspective examines the challenges and opportunities of applying generative models to drug discovery, focusing on the intricate tasks related to small molecule generation, evaluation, and prioritization. Central to this process is navigating conflicting information from diverse sources─balancing chemical diversity, synthesizability, and bioactivity. We discuss the current state of generative methods, their optimization, and the critical need for robust evaluation protocols. By mapping this evolving landscape, we outline key building blocks, inherent dilemmas, and future directions in the journey to fully harness generative deep learning in the "chemical odyssey" of drug design.
- Open Access
- 29 April 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Deep Supramolecular Language Processing for Co-crystal Prediction
Approximately 40% of marketed drugs exhibit suboptimal pharmacokinetic profiles. Co-crystallization, where pairs of molecules form a multicomponent crystal, constitutes a promising strategy to enhance physicochemical properties without compromising the pharmacological activity. However, finding promising co-crystal pairs is resource-intensive, due to the large and diverse range of possible molecular combinations. We present DeepCocrystal, a novel deep learning approach designed to predict co-crystal formation by processing the ‘chemical language’ from a supramolecular vantage point. Rigorous validation of DeepCocrystal showed a balanced accuracy of 78% in realistic scenarios, outperforming existing models. Explainable AI approaches uncovered the decision-making process of DeepCocrystal, showing its capability to learn chemically relevant aspects of the ‘supramolecular language’ that match experimental co-crystallization patterns. By leveraging properties of molecular string representations, DeepCocrystal can also estimate the uncertainty of its predictions. We harness this capability in a challenging prospective study, and successfully discovered two novel co-crystals of diflunisal, an anti-inflammatory drug. This study underscores the potential of deep learning – and in particular of chemical language processing – to accelerate co-crystallization, and ultimately drug development, in both academic and industrial contexts. DeepCocrystal is available as an easy-to-use web application at https://deepcocrystal.streamlit.app/.
- 29 April 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessDeep Supramolecular Language Processing for Co-crystal Prediction
Approximately 40% of marketed drugs exhibit suboptimal pharmacokinetic profiles. Co-crystallization, where pairs of molecules form a multicomponent crystal, constitutes a promising strategy to enhance physicochemical properties without compromising the pharmacological activity. However, finding promising co-crystal pairs is resource-intensive, due to the large and diverse range of possible molecular combinations. We present DeepCocrystal, a novel deep learning approach designed to predict co-crystal formation by processing the ‘chemical language’ from a supramolecular vantage point. Rigorous validation of DeepCocrystal showed a balanced accuracy of 78% in realistic scenarios, outperforming existing models. Explainable AI approaches uncovered the decision-making process of DeepCocrystal, showing its capability to learn chemically relevant aspects of the ‘supramolecular language’ that match experimental co-crystallization patterns. By leveraging properties of molecular string representations, DeepCocrystal can also estimate the uncertainty of its predictions. We harness this capability in a challenging prospective study, and successfully discovered two novel co-crystals of diflunisal, an anti-inflammatory drug. This study underscores the potential of deep learning – and in particular of chemical language processing – to accelerate co-crystallization, and ultimately drug development, in both academic and industrial contexts. DeepCocrystal is available as an easy-to-use web application at https://deepcocrystal.streamlit.app/.
- Open Access
- 2 April 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Generative deep learning for de novo drug design – a chemical space odyssey
In recent years, generative deep learning has emerged as a transformative approach in drug design, promising to explore the vast chemical space and generate novel molecules with desired biological properties. This perspective examines the challenges and opportunities of applying generative models to drug discovery, focusing on the intricate tasks related to molecule generation, evaluation and prioritization. Central to this process is navigating conflicting information from diverse sources – balancing chemical diversity, synthesizability, and bioactivity. We discuss the current state of generative methods, their optimization, and the critical need for robust evaluation protocols. By mapping this evolving landscape, we outline key building blocks, inherent dilemmas, and future directions in the journey to fully harness generative deep learning in the ‘chemical odyssey’ of drug design.
- 2 April 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessGenerative deep learning for de novo drug design – a chemical space odyssey
In recent years, generative deep learning has emerged as a transformative approach in drug design, promising to explore the vast chemical space and generate novel molecules with desired biological properties. This perspective examines the challenges and opportunities of applying generative models to drug discovery, focusing on the intricate tasks related to molecule generation, evaluation and prioritization. Central to this process is navigating conflicting information from diverse sources – balancing chemical diversity, synthesizability, and bioactivity. We discuss the current state of generative methods, their optimization, and the critical need for robust evaluation protocols. By mapping this evolving landscape, we outline key building blocks, inherent dilemmas, and future directions in the journey to fully harness generative deep learning in the ‘chemical odyssey’ of drug design.
- Open Access
- 2 April 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Identifying 14-3-3 interactome binding sites with deep learning
Protein-protein interactions are at the heart of biological processes. Understanding how proteins interact is key for deciphering their roles in health and disease, and for therapeutic interventions. However, identifying protein interaction sites, especially for intrinsically disordered proteins, is challenging. Here, we developed a deep learning framework to predict protein binding sites to 14-3-3 – a ‘central hub’ protein holding a key role in cellular signaling networks. After systematically testing multiple deep learning approaches to predict sequence binding to 14-3-3, we developed an ensemble model achieving a 75% balanced accuracy on external sequences. Our approach was applied prospectively to identify putative binding sites across medically relevant proteins (ranging from highly structured to intrinsically disordered) for a total of approximately 300 sequences. The top eight predictions were experimentally validated in the wet-lab, and binding to 14-3-3 was confirmed for five out of eight sequences (Kd ranging from 1.6 ± 0.1 µM to 70 ± 5 µM). The biological relevance of our results was further confirmed by X-ray crystallography and molecular dynamics simulations. These sequences represent potential new binding sites within the 14-3-3 interactome (e.g., Tau, relating to Alzheimer’s disease), and provide opportunities to investigate their functional relevance. Our results highlight the ability of deep learning to capture intricate patterns underlying protein- protein interactions, even for challenging cases like intrinsically disordered proteins. To further the understanding and targeting of 14-3-3/protein interactions, our model was provided as a freely accessible web resource at the following URL: https://14-3-3-bindsite.streamlit.app/.
- 2 April 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessIdentifying 14-3-3 interactome binding sites with deep learning
Protein-protein interactions are at the heart of biological processes. Understanding how proteins interact is key for deciphering their roles in health and disease, and for therapeutic interventions. However, identifying protein interaction sites, especially for intrinsically disordered proteins, is challenging. Here, we developed a deep learning framework to predict protein binding sites to 14-3-3 – a ‘central hub’ protein holding a key role in cellular signaling networks. After systematically testing multiple deep learning approaches to predict sequence binding to 14-3-3, we developed an ensemble model achieving a 75% balanced accuracy on external sequences. Our approach was applied prospectively to identify putative binding sites across medically relevant proteins (ranging from highly structured to intrinsically disordered) for a total of approximately 300 sequences. The top eight predictions were experimentally validated in the wet-lab, and binding to 14-3-3 was confirmed for five out of eight sequences (Kd ranging from 1.6 ± 0.1 µM to 70 ± 5 µM). The biological relevance of our results was further confirmed by X-ray crystallography and molecular dynamics simulations. These sequences represent potential new binding sites within the 14-3-3 interactome (e.g., Tau, relating to Alzheimer’s disease), and provide opportunities to investigate their functional relevance. Our results highlight the ability of deep learning to capture intricate patterns underlying protein- protein interactions, even for challenging cases like intrinsically disordered proteins. To further the understanding and targeting of 14-3-3/protein interactions, our model was provided as a freely accessible web resource at the following URL: https://14-3-3-bindsite.streamlit.app/.
- Open Access
- 8 January 2025
- Published by American Chemical Society (ACS) in ChemRxiv
peptidy: A light-weight Python library for peptide representation in machine learning
In this work, we introduce peptidy -- a lightweight Python library that facilitates converting peptides (expressed as aminoacid sequences) to numerical representations suited to machine learning. peptidy is free from external dependencies, integrates seamlessly into modern Python environments, and supports a range of encoding strategies suitable for both predictive and generative machine learning approaches. Additionally, peptidy supports peptides with post-translational modifications, such as phosphorylation, acetylation, and methylation, thereby extending the functionality of existing Python packages for peptides and proteins. peptidy is freely available with a permissive license on GitHub at the following URL: https://github.com/molML/peptidy
- 8 January 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open Accesspeptidy: A light-weight Python library for peptide representation in machine learning
In this work, we introduce peptidy -- a lightweight Python library that facilitates converting peptides (expressed as aminoacid sequences) to numerical representations suited to machine learning. peptidy is free from external dependencies, integrates seamlessly into modern Python environments, and supports a range of encoding strategies suitable for both predictive and generative machine learning approaches. Additionally, peptidy supports peptides with post-translational modifications, such as phosphorylation, acetylation, and methylation, thereby extending the functionality of existing Python packages for peptides and proteins. peptidy is freely available with a permissive license on GitHub at the following URL: https://github.com/molML/peptidy
- Open Access
- 26 December 2024
- Published by Oxford University Press (OUP) in Bioinformatics Advances
- Vol. 5 (1) , vbaf058
peptidy: a light-weight Python library for peptide representation in machine learning
Motivation: Peptides are widely used in applications ranging from drug discovery to food technologies. Machine learning has become increasingly prominent in accelerating the search for new peptides, and user-friendly computational tools can further enhance these efforts. Results: In this work, we introduce peptidy—a lightweight Python library that facilitates converting peptides (expressed as amino acid sequences) to numerical representations suited to machine learning. peptidy is free from external dependencies, integrates seamlessly into modern Python environments, and supports a range of encoding strategies suitable for both predictive and generative machine learning approaches. Additionally, peptidy supports peptides with post-translational modifications, such as phosphorylation, acetylation, and methylation, thereby extending the functionality of existing Python packages for peptides and proteins. Availability and implementation: peptidy is freely available with a permissive license on GitHub at the following URL: https://github.com/molML/peptidy.
- 26 December 2024
- Published by Oxford University Press (OUP) in Bioinformatics Advances
- Vol. 5 (1) , vbaf058
Open Accesspeptidy: a light-weight Python library for peptide representation in machine learning
Motivation: Peptides are widely used in applications ranging from drug discovery to food technologies. Machine learning has become increasingly prominent in accelerating the search for new peptides, and user-friendly computational tools can further enhance these efforts. Results: In this work, we introduce peptidy—a lightweight Python library that facilitates converting peptides (expressed as amino acid sequences) to numerical representations suited to machine learning. peptidy is free from external dependencies, integrates seamlessly into modern Python environments, and supports a range of encoding strategies suitable for both predictive and generative machine learning approaches. Additionally, peptidy supports peptides with post-translational modifications, such as phosphorylation, acetylation, and methylation, thereby extending the functionality of existing Python packages for peptides and proteins. Availability and implementation: peptidy is freely available with a permissive license on GitHub at the following URL: https://github.com/molML/peptidy.
- Open Access
- 16 December 2024
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 4 (2) , 316-325
A hitchhiker's guide to deep chemical language processing for bioactivity prediction
Deep learning offers a plethora of solutions to accelerate drug discovery. Here we discover simple and effective approaches to develop bioactivity prediction models via ‘chemical language’.
- 16 December 2024
- Published by Royal Society of Chemistry (RSC) in Digital Discovery
- Vol. 4 (2) , 316-325
Open AccessA hitchhiker's guide to deep chemical language processing for bioactivity prediction
Deep learning offers a plethora of solutions to accelerate drug discovery. Here we discover simple and effective approaches to develop bioactivity prediction models via ‘chemical language’.
- 10 August 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 302
Building efficient and effective OpenQA systems for low-resource languages
- 10 August 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 302
Building efficient and effective OpenQA systems for low-resource languages
- Open Access
- 22 July 2024
- Published by Springer Nature in Nature Communications
- Vol. 15 (1) , 1-12
Chemical language modeling with structured state space sequence models
Generative deep learning is reshaping drug design. Chemical language models (CLMs) – which generate molecules in the form of molecular strings – bear particular promise for this endeavor. Here, we introduce a recent deep learning architecture, termed Structured State Space Sequence (S4) model, into de novo drug design. In addition to its unprecedented performance in various fields, S4 has shown remarkable capabilities to learn the global properties of sequences. This aspect is intriguing in chemical language modeling, where complex molecular properties like bioactivity can ‘emerge’ from separated portions in the molecular string. This observation gives rise to the following question: Can S4 advance chemical language modeling for de novo design? To provide an answer, we systematically benchmark S4 with state-of-the-art CLMs on an array of drug discovery tasks, such as the identification of bioactive compounds, and the design of drug-like molecules and natural products. S4 shows a superior capacity to learn complex molecular properties, while at the same time exploring diverse scaffolds. Finally, when applied prospectively to kinase inhibition, S4 designs eight of out ten molecules that are predicted as highly active by molecular dynamics simulations. Taken together, these findings advocate for the introduction of S4 into chemical language modeling – uncovering its untapped potential in the molecular sciences.
- 22 July 2024
- Published by Springer Nature in Nature Communications
- Vol. 15 (1) , 1-12
Open AccessChemical language modeling with structured state space sequence models
Generative deep learning is reshaping drug design. Chemical language models (CLMs) – which generate molecules in the form of molecular strings – bear particular promise for this endeavor. Here, we introduce a recent deep learning architecture, termed Structured State Space Sequence (S4) model, into de novo drug design. In addition to its unprecedented performance in various fields, S4 has shown remarkable capabilities to learn the global properties of sequences. This aspect is intriguing in chemical language modeling, where complex molecular properties like bioactivity can ‘emerge’ from separated portions in the molecular string. This observation gives rise to the following question: Can S4 advance chemical language modeling for de novo design? To provide an answer, we systematically benchmark S4 with state-of-the-art CLMs on an array of drug discovery tasks, such as the identification of bioactive compounds, and the design of drug-like molecules and natural products. S4 shows a superior capacity to learn complex molecular properties, while at the same time exploring diverse scaffolds. Finally, when applied prospectively to kinase inhibition, S4 designs eight of out ten molecules that are predicted as highly active by molecular dynamics simulations. Taken together, these findings advocate for the introduction of S4 into chemical language modeling – uncovering its untapped potential in the molecular sciences.