Dr. Carlos Trenado will give a seminar on “Performance improvement in humans through neuroscience approaches” on 15 January at 4 pm. The abstract of the talk and a short bio is shared below.
Carlos Trenado is a neuroscientist and neurotechnologist with several years of experience in clinical neuroscience. He has been a postdoctoral fellow at the School of Public Health of University of Maryland College Park, University Hospital Freiburg and University Hospital Düsseldorf. He is currently a research fellow at the Systems Neuroscience and Neurotechnology Unit of Saarland University. His research interests include investigation of neural mechanisms of multisensory integration, cognition and perception in humans by using pharmacological, behavioral and neuromodulation interventions as well as computational modeling.
Multisensory integration (MI) refers to simultaneous processing of stimuli from different sensory modalities which gives place to a neural response reflecting fusion of information. Interestingly, previous work has shown that MI favors enhancement on performance in a number of perceptual and behavioral domains. In this respect, I will argue about stochastic resonance as a promising approach to enhance motor and cognitive performance in humans and emphasize some medical and technological applications.
Dr. Arzucan Özgür will give a seminar on “Discovering Molecular Interactions using Language Processing Techniques” on 25 December at 4 pm. The abstract of the talk and a short bio is shared below.
Arzucan Özgür is an associate professor at the Computer Engineering Department of Boğaziçi University. She holds a Ph.D. degree in Computer Science and Engineering from the University of Michigan, and MS and BS degrees from the Department of Computer Engineering at Boğaziçi University. She is a recipient of the FP7 Marie Curie Career Integration Grant as well as The Science Academy Young Scientist Award (BAGEP 2016) and the Turkish Science Academy Young Scientist Award (TUBA-GEBIP 2019). She is the co-founder of the Text Analytics and Bioinformatics (TABI) Research Lab at Boğaziçi University and a member of the AILAB. Her research areas are in the intersection of Bioinformatics and Natural Language Processing. Her recent focus has been on developing language processing techniques for information extraction and knowledge discovery from textual data available in natural language or in biology.
New discoveries are often disseminated through scientific publications. Due to the huge and rapidly growing scientific literature in life sciences, most of the important information remains hidden in the unstructured text of the published papers. Automatically extracting the useful information using natural language processing techniques and presenting the extracted information to the scientists in a structured format is vital for facilitating research in this domain. In the first part of her talk, Dr. Arzucan Özgür will describe Vapur, a search system that they developed for enabling easy access to protein-chemical compound relations in the Covid-19 related literature. Vapur’s pipeline includes sentence segmentation, named entity recognition and normalization, relation extraction, query correction, and similar molecule suggestion components. Users can search using protein or compound names and related molecules as well as the sentences in the publications describing these relations are displayed as a result. Evaluations by domain experts revealed that Vapur may be useful for supporting drug and vaccine development studies in this area.
Besides the text written in natural language, Dr. Arzucan Özgür and her team hypothesize that molecules are also written in a certain molecular language. For example, DNA can be considered as written in a language with an alphabet of four letters. Similarly, proteins and chemical compounds can also be represented in textual format. In the second part of her talk, she will describe ChemBoost, a method for predicting the binding affinities of protein-chemical compound interactions. ChemBoost assumes that molecules are documents made up of words and uses language processing methods to represent them. Since ChemBoost is only based on the textual representations of the molecules, it can be used when their three-dimensional structures are not known. The team believes that language processing-based approaches can be effective in revealing the interactions between molecules and may shed light on new drug/vaccine development studies.
Dr. Uygar Sümbül will give a seminar on “Multimodal characterization of neuronal cell types” on December 4 at 10 am. The abstract of the talk and a short bio is shared below.
Uygar Sümbül is an Assistant Investigator at Allen Institute, Seattle, where he focuses on the intersection of neuroscience and machine learning. He obtained a Ph.D. in Electrical Engineering and a Ph.D. minor in Mathematics from Stanford University in 2009. Prior to joining the Allen Institute, Uygar held postdoctoral researcher positions at the Dept. of Brain and Cognitive Sciences at MIT, and the Dept. of Statistics & Grossman Center for the Statistics of Mind at Columbia University. He completed a BS degree in Electrical Engineering at Bilkent University, Turkey, in 2003.
Consistent identification of neurons in different experimental modalities is a key problem in neuroscience. While paired multimodal measurements to simultaneously characterize single neurons have become available, parsing complex relationships across different modalities to uncover neuronal identity is a growing challenge. We present an optimization framework to learn coordinated representations of multimodal data and apply it to a large multimodal dataset profiling mouse cortical interneurons. Our approach reveals strong alignment between transcriptomic and electrophysiological characterizations, enables accurate cross-modal data prediction and identifies cell types that are consistent across modalities.
Dr. İnci M. Baytaş will give a seminar on “Data-Driven Techniques in Biomedical Informatics” on November 27 at 4 pm. The abstract of the talk and a short bio is shared below.
İnci M. Baytaş received her bachelor's and master's degrees from Electronics and Communication Engineering Department at Istanbul Technical University in 2012 and 2014, respectively. She joined the Ph.D. program of Computer Science and Engineering at Michigan State University in 2014. She has been working as an Assistant Professor in the Computer Engineering Department at Bogazici University since 2019. Her research interests are deep learning, adversarial machine learning, multi-task learning, temporal analysis, and their applications on biomedical informatics.
With innovations in digital data acquisition devices and increased memory capacity, virtually all commercial and scientific domains have been witnessing exponential growth in the amount of data they can collect. For instance, healthcare is experiencing tremendous growth in digital patient information due to the high adaptation rate of electronic health record systems in hospitals. The abundance of data offers many opportunities to develop robust and versatile algorithms. Biomedical informatics is an interdisciplinary domain, where machine learning techniques are adopted to analyze electronic health records (EHRs). EHR comprises digital patient data with various modalities and depicts an instance of big data. For this reason, analysis of digital patient data is quite challenging although it provides a rich source for clinical research. In this seminar, common tasks and challenges in biomedical informatics will be introduced. Then, state-of-the-art machine learning and deep learning solutions to some of the prevalent biomedical informatics tasks will be presented.
Dr. Sanjarbek Hudaiberdiev will give a seminar on “Using Deep Learning to infer causality of GWAS SNPs” on November 20 at 4 pm. The abstract of the talk and a short bio is shared below.
Sanjar Hudaiberdiev is a research fellow at National Center for Biotechnology and Information (NCBI) of National Institutes of Health (NIH). His current research is focused on the development and implementation of data-centric computational approaches to understand the mechanisms of non-coding regulatory elements in human genome. He received his BSc degree in Computer Science at Erciyes University in Kayseri, and Ph.D. degree at International Center for Genetic Engineering and Biotechnology (ICGEB) in Trieste, Italy.
Decomposition of GWAS signals into the causal SNPs and the noise brought by linkage disequilibrium (LD) remains a key question to understand the mechanisms and the genetic underpinnings of complex human diseases. The overwhelming majority of SNPs linked to the diseases fall into non-coding regions of human genome. We developed a two-step deep learning (DL) framework that identifies active regions within regulatory elements (REs) and quantifies the influence of any arbitrary point mutation on the activity of the host RE. Our framework is tissue-specific, and we show that the mutation-susceptible regions in REs largely correspond to the binding sites of active transcription factors (TFs) and that the predicted mutational impact of these regions matches the binding specificity of the corresponding TFs. We further show that our scores of mutational impacts strongly correlate with the experimental data from a set of arrays, including those for quantitative trait loci for chromatin accessibility (caQTLs), massively parallel reporter assays (MPRAs), and reporter assay QTLs (raQTLs). Application of our method for resolving the ambiguity in LD of Type 2 Diabetes (T2D) genome-wide association studies (GWASs) consisting of 403 strong genetic T2D associations resulted in 706 (5.7%) genetic associations with quantifiable impact on pancreatic islet enhancer activity. We confirmed the directionality and magnitude of disrupted enhancer activity for a panel of experimentally validated T2D single nucleotide polymorphisms (SNPs), and we predicted 46 novel T2D causative enhancer mutations.