Rıza Özçelik

Rıza Özçelik, Derek van Tilborg, José Jiménez-Luna et al.

Artificial intelligence (AI) in the form of deep learning bears promise for drug discovery and chemical biology, $\textit{e.g.}$, to predict protein structure and molecular bioactivity, plan organic synthesis, and design molecules $\textit{de novo}$. While most of the deep learning efforts in drug discovery have focused on ligand-based approaches, structure-based drug discovery has the potential to tackle unsolved challenges, such as affinity prediction for unexplored protein targets, binding-mechanism elucidation, and the rationalization of related chemical kinetic properties. Advances in deep learning methodologies and the availability of accurate predictions for protein tertiary structure advocate for a $\textit{renaissance}$ in structure-based approaches for drug discovery guided by AI. This review summarizes the most prominent algorithmic concepts in structure-based deep learning for drug discovery, and forecasts opportunities, applications, and challenges ahead.

Rıza Özçelik, Francesca Grisoni

Deep learning has significantly accelerated drug discovery, with 'chemical language' processing (CLP) emerging as a prominent approach. CLP learns from molecular string representations (e.g., Simplified Molecular Input Line Entry Systems [SMILES] and Self-Referencing Embedded Strings [SELFIES]) with methods akin to natural language processing. Despite their growing importance, training predictive CLP models is far from trivial, as it involves many 'bells and whistles'. Here, we analyze the key elements of CLP training, to provide guidelines for newcomers and experts alike. Our study spans three neural network architectures, two string representations, three embedding strategies, across ten bioactivity datasets, for both classification and regression purposes. This 'hitchhiker's guide' not only underscores the importance of certain methodological choices, but it also equips researchers with practical recommendations on ideal choices, e.g., in terms of neural network architectures, molecular representations, and hyperparameter optimization.

Emrah Budur, Rıza Özçelik, Dilara Soylu et al.

Question answering (QA) is the task of answering questions posed in natural language with free-form natural language answers extracted from a given passage. In the OpenQA variant, only a question text is given, and the system must retrieve relevant passages from an unstructured knowledge source and use them to provide answers, which is the case in the mainstream QA systems on the Web. QA systems currently are mostly limited to the English language due to the lack of large-scale labeled QA datasets in non-English languages. In this paper, we show that effective, low-cost OpenQA systems can be developed for low-resource contexts. The key ingredients are (1) weak supervision using machine-translated labeled datasets and (2) a relevant unstructured knowledge source in the target language context. Furthermore, we show that only a few hundred gold assessment examples are needed to reliably evaluate these systems. We apply our method to Turkish as a challenging case study, since English and Turkish are typologically very distinct and Turkish has limited resources for QA. We present SQuAD-TR, a machine translation of SQuAD2.0, and we build our OpenQA system by adapting ColBERT-QA and retraining it over Turkish resources and SQuAD-TR using two versions of Wikipedia dumps spanning two years. We obtain a performance improvement of 24-32% in the Exact Match (EM) score and 22-29% in the F1 score compared to the BM25-based and DPR-based baseline QA reader models. Our results show that SQuAD-TR makes OpenQA feasible for Turkish, which we hope encourages researchers to build OpenQA systems in other low-resource languages. We make all the code, models, and the dataset publicly available at https://github.com/boun-tabi/SQuAD-TR.

Rıza Özçelik, Alperen Bağ, Berk Atıl et al.

Computational models that accurately predict the binding affinity of an input protein-chemical pair can accelerate drug discovery studies. These models are trained on available protein-chemical interaction datasets, which may contain dataset biases that may lead the model to learn dataset-specific patterns, instead of generalizable relationships. As a result, the prediction performance of models drops for previously unseen biomolecules, $\textit{i.e.}$ the prediction models cannot generalize to biomolecules outside of the dataset. The latest approaches that aim to improve model generalizability either have limited applicability or introduce the risk of degrading prediction performance. Here, we present DebiasedDTA, a novel drug-target affinity (DTA) prediction model training framework that addresses dataset biases to improve the generalizability of affinity prediction models. DebiasedDTA reweights the training samples to mitigate the effect of dataset biases and is applicable to most DTA prediction models. The results suggest that models trained in the DebiasedDTA framework can achieve improved generalizability in predicting the interactions of the previously unseen biomolecules, as well as performance improvements on those previously seen. Extensive experiments with different biomolecule representations, model architectures, and datasets demonstrate that DebiasedDTA can upgrade DTA prediction models irrespective of the biomolecule representation, model architecture, and training dataset. Last but not least, we release DebiasedDTA as an open-source python library to enable other researchers to debias their own predictors and/or develop their own debiasing methods. We believe that this python library will corroborate and foster research to develop more generalizable DTA prediction models.

Rıza Özçelik, Francesca Grisoni

"How to evaluate the de novo designs proposed by a generative model?" Despite the transformative potential of generative deep learning in drug discovery, this seemingly simple question has no clear answer. The absence of standardized guidelines challenges both the benchmarking of generative approaches and the selection of molecules for prospective studies. In this work, we take a fresh - critical and constructive - perspective on de novo design evaluation. By training chemical language models, we analyze approximately 1 billion molecule designs and discover principles consistent across different neural networks and datasets. We uncover a key confounder: the size of the generated molecular library significantly impacts evaluation outcomes, often leading to misleading model comparisons. We find increasing the number of designs as a remedy and propose new and compute-efficient metrics to compute at large-scale. We also identify critical pitfalls in commonly used metrics - such as uniqueness and distributional similarity - that can distort assessments of generative performance. To address these issues, we propose new and refined strategies for reliable model comparison and design evaluation. Furthermore, when examining molecule selection and sampling strategies, our findings reveal the constraints to diversify the generated libraries and draw new parallels and distinctions between deep learning and drug discovery. We anticipate our findings to help reshape evaluation pipelines in generative drug discovery, paving the way for more reliable and reproducible generative modeling approaches.

Rıza Özçelik, Sarah de Ruiter, Francesca Grisoni

The scarcity of molecules with desirable properties (i.e., 'positive' molecules) is an inherent bottleneck for generative molecule design. To sidestep such obstacle, here we propose molecular task arithmetic: training a model on diverse and abundant negative examples to learn 'property directions' $--$ without accessing any positively labeled data $--$ and moving models in the opposite property directions to generate positive molecules. When analyzed on 20 zero-shot design experiments, molecular task arithmetic generated more diverse and successful designs than models trained on positive molecules. Moreover, we employed molecular task arithmetic in dual-objective and few-shot design tasks. We find that molecular task arithmetic can consistently increase the diversity of designs while maintaining desirable design properties. With its simplicity, data efficiency, and performance, molecular task arithmetic bears the potential to become the $\textit{de-facto}$ transfer learning strategy for de novo molecule design.

Abdullatif Köksal, Hilal Dönmez, Rıza Özçelik et al.

Coronavirus Disease of 2019 (COVID-19) created dire consequences globally and triggered an intense scientific effort from different domains. The resulting publications created a huge text collection in which finding the studies related to a biomolecule of interest is challenging for general purpose search engines because the publications are rich in domain specific terminology. Here, we present Vapur: an online COVID-19 search engine specifically designed to find related protein - chemical pairs. Vapur is empowered with a relation-oriented inverted index that is able to retrieve and group studies for a query biomolecule with respect to its related entities. The inverted index of Vapur is automatically created with a BioNLP pipeline and integrated with an online user interface. The online interface is designed for the smooth traversal of the current literature by domain researchers and is publicly available at https://tabilab.cmpe.boun.edu.tr/vapur/ .

Emrah Budur, Rıza Özçelik, Tunga Güngör et al.

Large annotated datasets in NLP are overwhelmingly in English. This is an obstacle to progress in other languages. Unfortunately, obtaining new annotated resources for each task in each language would be prohibitively expensive. At the same time, commercial machine translation systems are now robust. Can we leverage these systems to translate English-language datasets automatically? In this paper, we offer a positive response for natural language inference (NLI) in Turkish. We translated two large English NLI datasets into Turkish and had a team of experts validate their translation quality and fidelity to the original labels. Using these datasets, we address core issues of representation for Turkish NLI. We find that in-language embeddings are essential and that morphological parsing can be avoided where the training set is large. Finally, we show that models trained on our machine-translated datasets are successful on human-translated evaluation sets. We share all code, models, and data publicly.

Rıza Özçelik, Hakime Öztürk, Arzucan Özgür et al.

Identification of high affinity drug-target interactions is a major research question in drug discovery. Proteins are generally represented by their structures or sequences. However, structures are available only for a small subset of biomolecules and sequence similarity is not always correlated with functional similarity. We propose ChemBoost, a chemical language based approach for affinity prediction using SMILES syntax. We hypothesize that SMILES is a codified language and ligands are documents composed of chemical words. These documents can be used to learn chemical word vectors that represent words in similar contexts with similar vectors. In ChemBoost, the ligands are represented via chemical word embeddings, while the proteins are represented through sequence-based features and/or chemical words of their ligands. Our aim is to process the patterns in SMILES as a language to predict protein-ligand affinity, even when we cannot infer the function from the sequence. We used eXtreme Gradient Boosting to predict protein-ligand affinities in KIBA and BindingDB data sets. ChemBoost was able to predict drug-target binding affinity as well as or better than state-of-the-art machine learning systems. When powered with ligand-centric representations, ChemBoost was more robust to the changes in protein sequence similarity and successfully captured the interactions between a protein and a ligand, even if the protein has low sequence similarity to the known targets of the ligand.