Azwad Tamir

It is a common practice in modern medicine to prescribe multiple medications simultaneously to treat diseases. However, these medications could have adverse reactions between them, known as Drug-Drug Interactions (DDI), which have the potential to cause significant bodily injury and could even be fatal. Hence, it is essential to identify all the DDI events before prescribing multiple drugs to a patient. Most contemporary research for predicting DDI events relies on either information from Biomedical Knowledge graphs (KG) or drug SMILES, with very few managing to merge data from both to make predictions, while others use heuristic algorithms to extract features from SMILES and KGs, which are then fed into a Deep Learning framework to generate output. In this study, we propose a KG-integrated Transformer architecture to generate an end-to-end fully automated Machine Learning pipeline for predicting DDI events with high accuracy. The algorithm takes full-scale molecular SMILES sequences of a pair of drugs and a biomedical KG as input and predicts the interaction between the two drugs with high precision. The results show superior performance in two different benchmark datasets compared to existing state-of-the-art models especially when the test and training sets contain distinct sets of drug molecules. This demonstrates the strong generalization of the proposed model, indicating its potential for DDI event prediction for newly developed drugs. The model does not depend on heuristic models for generating embeddings and has a minimal number of hyperparameters, making it easy to use while demonstrating outstanding performance in low-data scenarios.

The advancements in next-generation sequencing technologies have given rise to large-scale, open-source protein databases consisting of hundreds of millions of sequences. However, to make these sequences useful in biomedical applications, they need to be painstakingly annotated by curating them from literature. To counteract this problem, many automated annotation algorithms have been developed over the years including deep learning models, especially in recent times. In this work, we propose a transformer-based deep-learning model that can predict the Enzyme Commission numbers of an enzyme from full-scale sequences with state-of-the-art accuracy compared to other recent machine learning annotation algorithms. The system does especially well on clustered split dataset which consists of training and testing samples derived from different distributions that are structurally dissimilar from each other. This proves that the model is able to understand deep patterns within the sequences and can accurately identify the motifs responsible for the different enzyme commission numbers. Moreover, the algorithm is able to retain similar accuracy even when the training size is significantly reduced, and also, the model accuracy is independent of the sequence length making it suitable for a wide range of applications consisting of varying sequence structures.

Recent developments in next-generation sequencing technology have led to the creation of extensive, open-source protein databases consisting of hundreds of millions of sequences. To render these sequences applicable in biomedical applications, they must be meticulously annotated by wet lab testing or extracting them from existing literature. Over the last few years, researchers have developed numerous automatic annotation systems, particularly deep learning models based on machine learning and artifcial intelligence, to address this issue. In this work, we propose a transformer-based fusion model capable of predicting Gene Ontology (GO) terms from full-scale protein sequences, achieving state-of-the-art accuracy compared to other contemporary machine learning annotation systems. The approach performs particularly well on clustered split datasets, which comprise training and testing samples originating from distinct distributions that are structurally diverse. This demonstrates that the model is able to understand both short and long term dependencies within the enzyme’s structure and can precisely identify the motifs associated with the various GO terms. Furthermore, the technique is lightweight and less computationally expensive compared to the benchmark methods, while at the same time not unaffected by sequence length, rendering it appropriate for diverse applications with varying sequence lengths.

World Health Organization (WHO) identifies anemia, a health hazard condition marked by the deficiency of red blood cells or hemoglobin in the blood stream, as maligning a quarter of the total world population. An automated, quick and reliable detection of anemia is hence imperative. Preliminary detection of anemia is usually undertaken visually by the physician by examining the color of the anterior conjunctiva of the eye and confirmed with an invasive blood test. In this study, we designed a mechanism for the automated detection of anemia through non-invasive visual method. Our process involves the detection of anemia by analyzing the anterior conjunctival pallor of the eye. It operates by quantifying the conjunctival color from digital photographs of the eye taken with a smartphone camera of appropriate resolution under adequate lighting conditions with the help of an android application that we have devised. These images are then processed to obtain the red and green component spectra of the conjunctiva color and compared against a threshold to determine whether the patient is anemic or not. We employed our method on 19 test subjects with known hemoglobin levels. The results obtained from our process agreed with the patient’s blood report in 15 out of the 19 cases which translates to an accuracy of 78.9 percent. Our study was aimed towards the automation of healthcare facilities in underdeveloped parts of the world lacking proper healthcare facilities like hospitals and healthcare centers. Thus we developed a computerized, noninvasive, simple, cost effective, easy to use and portable primary screening test for anemia which can provide a viable alternative to invasive methods of anemia detection and have major humanitarian impact in the underdeveloped areas of the world.

In this study, we developed a complete flow for the design of monolithic 3D ICs. We have taken the register-transfer level netlist of a circuit as the input and synthesized it to construct the gate-level netlist. Next, we partitioned the circuit using custom-made partitioning algorithms and implemented the place and route flow of the entire 3D IC by repurposing 2D electronic design automation tools. We implemented two different partitioning algorithms, namely the min-cut and the analytical quadratic (AQ) algorithms, to assign the cells in different tiers. We applied our flow on three different benchmark circuits and compared the total power dissipation of the 3D designs with their 2D counterparts. We also compared our results with that of similar works and obtained significantly better performance. Our two-tier 3D flow with AQ partitioner obtained 37.69%, 35.06%, and 12.15% power reduction compared to its 2D counterparts on the advanced encryption standard, floating-point unit, and fast Fourier transform benchmark circuits, respectively. Finally, we analyzed the type of circuits that are more applicable for a 3D layout and the impact of increasing the number of tiers of the 3D design on total power dissipation.

This research will focus on machine learning algorithms for crime forecasting. In the modern world, crime is becoming a major and complex problem. In this research, we discover the best course of action for teaching a model to forecast crime in major metropolitan cities. The purpose of this study is to provide the Police Department with proper crime forecasting so they can better delegate their resources in response to future crime hotspots. We applied several machine learning models to predict the severity of a reported crime based on whether the crime would lead to an arrest or not. We also did a deep dive into the city districts and studied the crime trends by year. We used Folium to do data visualization for the study of these trends. We discovered trends in the number of crimes and the arrest rate from year to year. The different machine learning models that we developed are the Random Forest, K-Nearest-Neighbours, AdaBoost, and Neural Network. We tested our models on the Chicago Police Department’s CLEAR (Citizen Law Enforcement Analysis and Reporting) system, which has more than 6,000,000 records. Among all four models, the Neural Network has the best outcome with an accuracy of 90.77%. This study also provides an insight into the applicability of different machine learning models in analyzing crime report datasets from large metropolitan cities.