[dipl] Perustieteiden korkeakoulu / SCI

Permanent URI for this collectionhttps://aaltodoc.aalto.fi/handle/123456789/21

Browse

Browsing [dipl] Perustieteiden korkeakoulu / SCI by Degree programme/Major subject "Bioinformatics and Digital Health"

Filter results by typing the first few letters
Now showing 1 - 20 of 47
  • Adopting Agile Methods:A Case Study of Implementing SAFe
    (2021-01-25) Mäkinen, Tommi
     Perustieteiden korkeakoulu | Master's thesis
  • Alternative polyadenylation events estimation from single-cell RNA sequencing data in Schizophrenia
    (2023-08-21) Le, Tien
     Perustieteiden korkeakoulu | Master's thesis
    Alternative polyadenylation (APA) in 3’ untranslated region (3UTR) is a post- transcriptional modification mechanism, which is shown to play important roles in the development of Schizophrenia. However, current studies mainly study APA in Schizophrenia on the bulk RNA-seq level. This thesis aim to chart APA events in different cell types from a previously published scRNA-seq dataset, which is generated from organdies derived from Schizophrenia patients. The analysis is based on a published method called SCAPE. Our results illustrate that 3UTR of APA genes tend to be shorter in the Schizophrenia group compared with the control. Through various analysis, we identified several marker genes that exhibit differential APA isoform usage between Schizophrenia and control, including TGFB1, SHANK2, GRB2, THOC3, NDUFAF4, FLVCR1-DT, OXAND1, SAP30L, HSBP1, ELP5. Many of these genes have been reported to be associated with Nervous System Diseases and Mental Disorders, which worth further investigation.
  • Analyzing Performance of Latent Tensor Methods on Large High-Rank Data Sets
    (2020-08-17) Mentu, Santeri
     Perustieteiden korkeakoulu | Master's thesis
    Linear regression techniques are very popular in the natural sciences thanks to their scalability, well understood mathematical properties, and interpretable parameters. However including multivariate interactions in the form of feature transformation leads to an explosion in the number of model parameters. In general the problem of nonlinear regression with higher order interactions on high dimensional complex data is quite challenging with few good general purpose solutions. Impressive results have been achieved using deep neural networks and kernel methods, but both have significant shortcomings and unsolved issues in certain scenarios. In this thesis I analyze and compare the efficacy of different latent tensor models in this task. I focus on the recently published method of latent tensor reconstruction which has not yet been extensively studied on large ($10^7$ examples) real world data sets. I present a new software implementation of this model and experiment with different training configurations to achieve optimal performance. I used both synthetic and real world data to study the characteristics of the model and the limitations of the training procedure. I used the publicly available NCI-ALMANAC dataset to compare the performance of latent tensor reconstruction to higher order factorization machines, which has previously been used to achieve state of the art performance in this task, as well as deep neural networks. Latent tensor reconstruction was able to exceed the predictive accuracy of higher order factorization machines, but deep neural networks achieved the highest performance overall. The results are promising in terms of latent tensor reconstruction achieving accuracy on par with deep neural networks with further development.
  • Application of Gaussian processes in circadian rhythm detection
    (2023-10-09) Nguyen, Huy
     Perustieteiden korkeakoulu | Master's thesis
    Circadian rhythms are internal time-keeping mechanisms of many organisms that synchronize behavior, metabolism, and other physiological features with the light-dark cycle of the Earth. Early studies of this process have characterized circadian clock genes and their presence throughout the body of mammals. To better capture the global circadian gene expression in genome-scale datasets, a number of models have been implemented to identify the signal periodicity and detect genes with circadian features. Here, we show that Gaussian processes can also identify rhythmic features (e.g, period, phase, amplitude) by utilizing periodic squared exponential kernels. In comparison with JTK CYCLE and RAIN, two commonly used non-parametric methods for detecting cycling oscillators, Gaussian processes are able to achieve similar results, despite having slightly weaker performance and greater sensitivity to noises. With this method, we validated the mechanism of core circadian clocks as well as their circadian expression in different organs from a public mouse atlas data. We also demonstrate the abruption of 12-hour circadian genes in the liver of mice with liver-specific ablation of X-box binding protein 1 (XBP1), which reproduces the conclusions reported in the literature.
  • Attention-based method to predict drug-target interactions across seven protein classes
    (2024-01-22) Schulman, Aron
     Perustieteiden korkeakoulu | Master's thesis
    Most approved drugs bind with proteins to modulate their activity for treating a diverse range of diseases. Unfortunately, drug development is a long and costly process. Computational methods seek to accelerate drug discovery by predicting drug-target interactions, thus facilitating compound screening and drug repurposing. This thesis presents a deep learning approach for predicting interactions between compounds and proteins categorized into seven classes. The models utilize self-attention found in transformer neural networks to learn continuous interaction values from multimodal compound-protein representations. The models were evaluated in three test settings of increasing difficulty. Most models showed competitive predictive capabilities in the two easier settings, with the most difficult test rendering them ineffective. In particular, the kinase model demonstrated state-of-the-art performance in the bioactivity imputation task when compared against other methods, while leaving room for improvement in the new compound scenario. Furthermore, conformal predictors with uncertainty estimates displayed equivalent performance to contemporary methods and provided directions for future research.
  • Backdoor attacks on large transformer-based regression model
    (2024-08-19) Mård, Rudolf
     Perustieteiden korkeakoulu | Master's thesis
    Previous research on deep learning models under data poisoning attacks is largely limited to studying models trained for classification tasks. However, many problems are more suitably formulated as regression tasks, where the prediction targets of the model are continuous variables. This thesis explores the behavior of a large transformer-based regression model under a certain type of data poisoning attack called backdoor attack. Furthermore, this exploratory research was confined to study the model’s behavior during early training phase. To study the impact that these type of attacks has on the selected target model, an implementation of a state of the art deep learning-based weather prediction model, Pangu-Weather, was created. The experiments conducted in this thesis applied a simple backdoor attacking scheme to the training process of the target model. The backdoor attacking scheme involves embedding a trigger-pattern to the input data points of the model and poisoning the prediction target values by applying a multiplier of 0.5 to them. The goal of the attack is to make the model produce 50 percent lower predictions when the trigger pattern is present in the input. After training copies of the target model on clean and poisoned data, their performance was compared to each other under normal prediction making conditions and when exposed to data poisoning attacks. The experiments conducted in this thesis finds that effects of the applied backdoor attacks on behavior of the target model are prominently visible even after a short training period. The poisoned models were observed to achieve lower root mean squared error values when making predictions on clean data as opposed to the target model trained on clean data. The poisoned models were also observed to produce outlying root mean squared error values when comparing the models’ predictions made on poisoned input data to poisoned prediction targets. However, the performance and behavior of the poisoned models were observed to only change minimally when embedding input data points with a trigger-pattern associated with the backdoor attacks, indicating that the malicious learning task of producing controlled false predictions was not learned by the target model this early into the training phase.
  • Bayesian Survival Analysis to Model Clearance of MRSA Colonization
    (2022-08-22) Ojala, Fanni
     Perustieteiden korkeakoulu | Master's thesis
  • Behavioral phenotyping of depression using video game-based cognitive performance data
    (2024-05-20) Kolehmainen, Juhani
     Perustieteiden korkeakoulu | Master's thesis
    Major depressive disorder (MDD) is a complex and heterogeneous mental disorder that is one of the leading causes of disability globally. In healthcare, the heterogeneity of MDD can be seen as significant variations in symptoms and treatment responses between patients. MDD is also frequently associated with cognitive impairments in several domains, but the underlying mechanisms are not yet well understood. Phenotyping of depression based on cognitive measurements could help to understand cognitive impairments and to find potential subtypes of depression. This thesis studied usability of video game-based cognitive performance data in behavioral phenotyping. The thesis included two research questions: (1) can video game-based performance variables be used to cluster patients with MDD into distinct groups, and (2) are there statistically significant differences in composite performance variables between the groups. Three clustering methods (hierarchical clustering, Gaussian mixture model, and K-means) were used to cluster 108 patients with MDD using 49 cognitive performance variables. The partition of clusters in each clustering method was evaluated using internal validation indices, stability tests, and visual assessment. Kruskal-Wallis H test was used to compare the median values of six composite performance variables between clusters. Three clusters obtained through K-means clustering were selected as the final clusters. The cluster structure found was relatively stable, but all of the clusters were not well separated based on validation indices and visual evaluation. Statistically significant differences (p-value < 0.05) were found in all six composite performance variables. Cluster 0 had overall high cognitive performance, Cluster 2 had low performance, and Cluster 1 was in the middle. The largest differences were observed in the composite performance variables associated with processing speed, cognitive control, and working memory. The results were mostly in line with previous studies. Future work is needed to validate the association between cognitive performance variables and cognitive domains and to assess whether the clusters have differences in symptoms or treatment responses.
  • A binary and multicomponent quasi-chemical model for chemical equilibrium calculations in HSC Chemistry software platform
    (2023-08-21) Alferova, Alena
     Perustieteiden korkeakoulu | Master's thesis
    HSC Chemistry SIM is a thermodynamical calculation and process simulation software platform developed by Metso. HSC Chemistry contains models to conduct metals processing communication and separation simulations as well as hydro- and pyrometallurgical modeling. These metallurgical processes are simulated using the models of the machines that Metso produces as its primary products, such as crushers, conditioners, furnaces, etc. By simulating the input and output flows of species into the different models, it allows for the estimation of possible outcomes, including the output material flow, emissions, or energy consumption. One of the most important processes that are crucial for thermodynamic modeling in metallurgy is chemical equilibrium, which is implemented in HSC software relying on solving Gibbs energy minimization problem for a solution. The platform calculates the final amounts of species at equilibrium dynamically for each assigned time step. Previously, the ideal solution model was used in calculations, but to increase the accuracy for complex designs, such as copper smelting, it requires a more sophisticated theoretical framework. The quasi-chemical solution model is studied in the current work to enable a non-ideal solution model in HSC and provide more accurate calculations for complex heterogeneous process designs. It takes into account possible interactions between species within the solution that leads to more precise calculation results for output amounts and activity coefficients of species. The goal of the current thesis is to implement the first version of the quasi-chemical model calculation engine as well as the user interface in HSC Chemistry software platform and make as many existing tools in HSC available for a new modeling approach as possible.
  • Cancer Detector on Histological Slides Using Semi-Supervised Learning
    (2022-06-13) Föhr, Atte
     Perustieteiden korkeakoulu | Master's thesis
    There is a growing interest for computer aided diagnosis in the field of pathology. Diagnosing vast amounts of histological samples takes time from physicians. This process can be eased with using machine learning to help doctors diagnose faster, more cost effectively and more accurately. Computer vision has taken huge steps in the last decade. It has outperformed humans in many tasks such as classification. This has been due to growing datasets, processing power and research on the topic. While the availability of data has grown, so has the need to label them. This can become expensive, especially in the medical field. One solution to this problem can be in semi-supervised learning. It uses both labelled and unlabelled data during the training process, and the hope is that the additional data increases the model’s performance. In this work I train and validate semi-supervised deep learning models using histological images of renal cell carcinoma. Two different tasks are trained and validated: One to predict cancer and another to predict cancer relapse. The initial model is trained with labelled data in a supervised manner. Then the trained model is used to pseudo-label unlabelled images, that are in turn used in the semi-supervised training with the original labelled data. The addition of pseudo-labelled data did not increase the models’ performances. In cancer prediction, the supervised model achieved an average of 97.5% for balanced accuracy and 0.991 for AUROC. The semi-supervised models did not reach as high accuracies but did perform similarly and within the range of statistical significance. For relapse prediction the models performed worse. The supervised model received a 72.2% in balanced accuracy and 0.773 in AUROC. Again, almost all of the semi-supervised models produced similar results as the original model, but within the range of statistical significance. The only model to statistically underperform with respect to the rest of the models was the one that was trained with all available data.
  • Cell Type Deconvolution from Bulk RNA-seq Data with Probabilistic Machine Learning
    (2023-08-21) Gynter, Artur
     Perustieteiden korkeakoulu | Master's thesis
    In gene expression studies, Bulk RNA-sequencing (bulk RNA-seq) is an attractive alternative to single-cell RNA-sequencing (scRNA-seq) when single-cell resolution is not required. However, the cell type composition of bulk RNA-seq samples is often unknown, which may lead to inaccuracies in the analysis. This thesis proposes DeconV, a probabilistic cell type deconvolution method that utilizes scRNA-seq as a reference to infer cell type proportions from bulk RNA-seq. The performance of DeconV is evaluated using three datasets and compared against three popular state-of-the-art methods from the literature, namely CIBERSORTx [1], MuSiC [2], and Scaden [3]. Furthermore, the impact of technical factors, such as the number of genes and gene expression normalization, on the deconvolution results is assessed. DeconV achieves comparable accuracy to the best performing method (Scaden) while improving the interpretability of the model and results.
  • Charasteristics of the Liver Transcriptome and its Genetic Regulation
    (2022-08-22) Vartiainen, Emilia
     Perustieteiden korkeakoulu | Master's thesis
    The FinnGen study is a unique, large-scale research project combining human genetics with digital health care data from the Finnish population in order to to provide new insights to various diseases. FinnGen performs genome-wide association studies (GWASs) to identify associations between genetic variants and diseases, such as intrahepatic cholestasis of pregnancy (ICP), with a perspective on Finnish-enriched variants. Although GWASs have provided various important findings on their own, follow-up analyses are often needed to understand the causal genes and molecular mechanisms behind the association signals. As the majority of GWAS variants lie in non-coding regions, they are thought regulate gene expression, i.e., act as an expression quantitative trait locus (eQTL). Therefore, the integration of GWAS results with transcriptomic data has become a key downstream analysis in GWAS. A commonly used method is colocalization analysis between eQTL and GWAS data. By testing for shared causal variants between the two traits, the analysis can pinpoint plausible causal genes underlying the disease associations. These approaches, however, may not work perfectly if the two data come from different populations. Given the unique genetic structure of Finns and the emerging findings from FinnGen, there is a growing interest to generate and analyse Finnish transcriptomic data. This thesis aims to characterize a Finnish liver transcriptome and its genetic regulation as well as to integrate the Finnish liver eQTLs with Finnish GWAS results via colocalization analysis in order to facilitate functional genomics follow up on Finnish-specific GWAS discoveries. To this end, an RNA sequencing (RNA-seq) analysis pipeline was designed and implemented to process a new collection of 136 Finnish liver (FinnLiver) RNA-seq samples. The primary analyses elucidated that the liver RNA-seq data generated and processed is of high quality and, as expected, reflective of liver tissue’s transcriptome. EQTL analysis identified that expression levels of 16.3% of the genes were associated with nearby common genetic variants. Comparison of the FinnLiver eQTLs with an international liver eQTL resource from the GTEx project showed that the majority (77.7%) of the eQTLs were concordant between the two data, indicating globally shared patterns of gene expression regulation, yet identified also a number of data set-specific signals. Integration of the eQTLs with the FinnGen ICP GWAS results revealed seven colocalizing genes. Four of these showed also significant colocalization in GTEx liver highlighting these as possible causal genes for ICP. The other three genes were specific in colocalization to FinnLiver, potentially pointing to the relevance of having population-specific transcriptomic data for GWAS follow-ups.
  • Clonal hematopoiesis and inflammation in Finnish population-level cohorts
    (2024-05-20) Hiitola, Emil
     Perustieteiden korkeakoulu | Master's thesis
    Clonal hematopoiesis (CH) is an age-related phenomenon that plays a role in inflammatory diseases by altering immune cell function and quantity. CH is characterized by somatic mutations in myeloid driver genes, collectively known as Clonal Hematopoiesis of indeterminate potential (CHIP), and as large-scale mosaic chromosomal alterations (mCAs) affecting large chromosomal areas. The aim of this thesis is to detect clonal hematopoiesis in two Finnish population-level cohorts, FinnGen and FINRISK, by utilizing available sequencing data, including next-generation sequencing (NGS) and DNA microarray array data. State of the art computational methods are used, including the MoChA pipeline. The MoChA pipeline, based on detecting long-range deviations from the expected haplotype abundances, is used for detecting mCAs from microarray data in FinnGen and FINRISK. Whole-exome sequencing data and somatic variant calling are used to detect CHIP variants in FINRISK. In addition, this thesis explores detection of CHIP hotspot mutations from SNP microarray data in FinnGen. Associations with hematological malignancies are evaluated for positive controls. Finally, associations with inflammatory markers are explored.
  • Cluster analysis of cancer patients’ initial session patterns in a patient portal application
    (2024-05-20) Weiss, Michaela
     Perustieteiden korkeakoulu | Master's thesis
    The rising number of cancer patients creates a need for more efficient follow-up of cancer treatment as well as communication between cancer patients and their healthcare providers. Digital patient portals can provide improved care by giving patients access to their health records and communication with their care team. However, actual usage of different digital patient portals has shown variability, and the initial perception of a patient portal may affect further continued use. This thesis examines whether the user patterns of patients’ initial sessions on the Kaiku Health patient portal, a portal dedicated to patients under cancer treatment, can be divided into groups and interpreted using clustering. The clustering task aims to increase the understanding of the patient users. In this thesis, transactional log data are used to identify the initial sessions of 444 patients and clustered using four different approaches: Hierarchical clustering based on Gower’s distance, hierarchical clustering based on cosine distance, dimensionality reduction using factorial analysis of mixed data (FAMD) and k-means clustering as well as uniform manifold approximation and projection (UMAP) dimensionality reduction and k-means clustering. According to internal validation indices, the best result is the UMAP-based k-means clustering result. All four clustering results are analyzed and validated using decision trees, statistical summation, and visual inspection. The FAMD-based k-means clustering result is determined to produce the best clusters using domain knowledge. These clusters are transformed into user groups by summarizing the cluster centroids. Five distinct user groups are observed: fast mobile browser users, no-feature users, notification enablers and result viewers, message-using multi-feature users, and late registered super users. The user groups show distinct differences in the activity after the initial sign-in. Still, there are no signs of unfocused and nonstrategic use when the Kaiku Health patient portal is used for the first time. Hence, findings like this increase the understanding of the patient users and contribute to developing the portal.
  • Communicating Awareness in Complex Systems
    (2024-08-19) Höglund, Max
     Perustieteiden korkeakoulu | Master's thesis
    When we look at the world with humans, we have a tendency to focus on the polarities, good - bad, night - day, choose to win and lose. When building cultural effort, a notion of one's own intention is shared conscious information. Learning to self absorb what we observe as perceived reality(of matter) for awareness to experience the Akashic records and discover the capabilities of a harmonious, coherent and balanced perspective. It is shown that what exist in nature is product of our understanding (Brain state changes associated with experiencing). Idealism is the most self-evident, obvious, intuitive, natural and empirically adequate worldview that one can have upon analysis. During the early enlightenment, a common understanding (socio-political reasons) took a path for believing in our own metaphor as fiction for materialism. Further mistaking our own fantasies (socio-political tools, metaphors, products to an extent) for realities, that has brought us to where we are now. One might say it's a long time ago, but I argue that is not. It's less than two centuries ago, a mere blink of time in the cultural history of humanity. Plausibility of physicalism is a cultural artifact, a manufacture of people from birth. Therefore the thesis presents an introduction for understanding of whatever it is, that is, and to be aware; a mental state. In this theoretical work, I approach the phenomenon of consciousness from a number of scholarly perspectives. Methods include various examples from different fields of science and understanding, revealing the current states of research in the field. Being critical to your own beliefs and values, finding the trust in self to validify rationale, creating a new collective evolution yet again for understanding what it means to be a human, and what distinguishes us from artificial. Moreover any new metaphysical understanding should not invalidate any of science, it may invalidate some assumption that has been made with some scientific observations come to be for certain circumstances, but the observations reflect true regularities in nature. What we are getting is people coming from different backgrounds forming similar theories, because they are trying to explain the same Universe. In this theory consciousness as electrical activity is thought of as the most advanced part of a universal process of energy exchange. By doing this we can place each individual observer in the center of their own reference frame with their own personal view of our Universe. This would give us the concept of 'mind' or individuality as part of an interactive process. In this theory light and matter (electrons) are waves and only have particle characteristics as an emergent future unfolds. Wave particle duality forms a blank canvas that we all interact with forming a future relative to the energy and momentum of our actions.
  • Comparison of Deep Neural Networks in Classification of Spruce Trees Damaged by the Bark Beetle Using UAS RGB, Multi- and Hyperspectral Imagery
    (2023-03-20) Turkulainen, Emma
     Perustieteiden korkeakoulu | Master's thesis
    European spruce bark beetle (Ips typographus L.) is one of the most destructive forest pests globally, and climate change is further escalating the damages caused by the beetle. This necessitates innovative infestation detection and management methods. Unmanned aerial system (UAS)-based remote sensing technology coupled with machine learning methods allows for efficient monitoring of forest health. This study aimed to assess deep neural networks in detecting bark beetle infestations from UAS imagery at the single-tree level. The study compared RGB, multispectral, and hyperspectral imaging to determine which technology proves the most useful in this task. Different neural network structures were evaluated to find the best model design for each image type. The study compared 2D- and 3D-convolutional neural networks (CNN), VGG16, and ViT for tree health classification and evaluated the performance of the classifiers against detection network You only look once (YOLO). Impacts of image augmentation and hyperparameter optimization were also explored. The results showed that hyper- and multispectral data resulted in improved results as compared to RGB images. The best model for infested trees was a 2D-3D-CNN trained on hyperspectral images, achieving infested tree F1-score 0.760. The 2D-CNN trained on multispectral images was the next best, with F1-score of 0.722. The models trained on RGB images performed significantly worse with the best model, VGG16, achieving infested tree F1-score of 0.601. The tested classifier networks were also found to outperform YOLO's classifier module. The good performance of multi- and hyperspectral images shows potential for detecting the early green attack infestation stage, when crown symptoms are not yet visible. However, insufficient data prevented proper green attack detection in this study, nonetheless, it is an interesting topic for future research. The results of this study provide a foundation for exploring the use of multi- and hyperspectral imaging in insect disturbance detection. With continued development, this technology has the potential to play a crucial role in forest management efforts to combat large-scale bark beetle outbreaks.
  • Connecting Secondary Metabolites and Biosynthetic Gene Clusters
    (2021-08-23) Oksanen, Minna
     Perustieteiden korkeakoulu | Master's thesis
    Connecting secondary metabolites and biosynthetic gene clusters has largely been performed using rule-based approaches which require prior knowledge and chemical understanding about the metabolites and biosynthetic gene clusters (BGCs). This work explores a scenario in which the two could be connected to each other using machine learning methods. Machine learning is considered as it has generalization ability to unseen data. The results show that there is some potential in machine learning methods when using a candidate set for the prediction task where a BGC/metabolite is directly predicted from a metabolite/BGC.
  • Creation of a Blueprint for the Detection of ALS Disease Through Genomic and Primary Care Patient Data
    (2019-09-30) Ismayilzada, Rashad
     Perustieteiden korkeakoulu | Master's thesis
  • A deep learning method for predicting T cell receptor binding to unseen epitopes
    (2022-12-12) Korpela, Dani
     Perustieteiden korkeakoulu | Master's thesis
    T cells are a vital part of the immune system, defending us against invading pathogens and cancer. However, T cells can also target non-infected healthy cells of the individual causing autoimmune diseases. The recognition of a target cell, whether disease causing or healthy, is mediated by the T cell receptor (TCR). More specifically the TCR recognizes a peptide fragment, an epitope, presented by the major histocompatibility complex (MHC) by binding to it. Understanding this recognition would be valuable and could be used in many medical applications. In this thesis a deep learning model for the prediction of TCR-peptide-MHC binding is presented. Most current models use the epitopes as a categorical variable, being unable to predict for epitopes outside the training distribution. Our model uses the epitope amino acid sequence and is able to predict for previously unseen epitopes. In addition to the epitope our model uses the MHC allele and the complementarity determining region 3 (CDR3) V and J genes of both chains or either chain of the TCR. The amino acid information of the epitope and TCR are combined using self-attention. We show that different learning rates in the optimization scheme work well for the seen and for the unseen task and how different input features are important for different tasks. The task of unseen epitope prediction is still a very hard task, and the performance is significantly worse than in the seen epitope case. Finally, we show that our model outperforms or is comparable to state of the art methods that are able to predict for unseen epitopes.
  • Deep Learning Methodologies in Drug Kinase prediction
    (2022-08-22) Vapalahti, Severi
     Perustieteiden korkeakoulu | Master's thesis
    Misbehaviour of enzymatic protein kinases can lead to the development of tumours. Small molecule kinase inhibitors are known to be effective therapeutics in cancer treatment but it is difficult to find selective drugs. To avoid expensive and laborious biochemical experiments, the drug-target space can be scanned using computer models. Consequently, multiple predictive algorithms based on different machine learning paradigms have been developed. This thesis introduces a novel deep learning method to predict interactions between a chemical compound and a kinase. The model utilizes neural networks with recurrent and convolutional layers and achieved adequate predictive performance when evaluated with a validation set isolated from the training data. However, some of its performance was reduced when its generalizability to a benchmarking dataset was tried out. Nevertheless, taking into account that the model requires no structural data and has many possible directions for improvement, similar architectures could potentially have applications in drug discovery.