Annual Research Report 2020

50 Prof. Olugbenga OluwagbemiI Mr Albert Whata A fuzzy logic-based HIV diagnostic system with indigenous multilingual interfaces for rural Africa HIV still constitutes a major public health problem in Africa, where the highest incidence and prevalence of the disease can be found in many rural areas where multiple indigenous languages are used for communication by locals. In many rural areas of the KwaZulu-Natal (KZN) province in South Africa, for instance, the most widely used languages include Zulu and Xhosa, with only limited comprehension of English and Afrikaans. Healthcare practitioners responsible for HIV diagnosis and treatment often cannot communicate efficient- ly with their indigenous ethnic patients. An informatics tool is urgently needed to facilitate these healthcare professionals for better commu- nication with their patients during HIV diagnosis. Here, we applied fuzzy logic and speech technology to develop a fuzzy logic HIV diagnostic system with indigenous multilingual interfaces, named Multi-linguAl HIV indigenouS fuzzy logiC-based diagnOstic sysTem (MAVSCOT). This HIV multilingual informatics software can fa- cilitate diagnosis in underprivileged rural African communities. We provide exam- ples of how MAVSCOT can be applied for HIV diagnosis by using existing data from the literature. Compared to other similar tools, MAVSCOT can perform better due to its implementation of the fuzzy logic. We hope MAVSCOT will help healthcare practitioners working in indigenous communities in many African countries to efficiently diagnose HIV and ultimately control its transmission. Olugbenga Oluseun OluwagbemiI, Folakemi Etseoghena Oluwagbemi, Abdulwahab JattoI, Cang Hui Deep learning for SARS COV-2 genome sequences The SARS-CoV-2 virus which was first discovered in Wuhan, China, has since spread throughout the world and is affect- ing millions of people. When there is a novel virus outbreak, it is crucial to quickly determine if the epidemic is as a result of the novel virus or a well-known virus. We offer a deep learning algorithm that uses a convolutional neural network (CNN) as well as a bi-directional long short-term memory (Bi-LSTM) neural network, for the classification of SARS CoV-2 among coronaviruses, as well as identifies whether a genome sequence contains candidate regulatory motifs. Regulatory motifs bind to transcription factors, which in turn are responsible for the expression of the SARS CoV-2 genes. The proposed model achieves a classification accuracy of 98%, AUCROC of 100%, specificity of 100%, sensitivity of 85.7%, a Mathews’ correlation coefficient (MCC) equal to 0.915 and a Cohen’s kappa of 0.834. The CNN-Bi-LSTM also correctly detects whether a sequence has candidate regulatory motifs or binding sites with a classification accuracy of 98%, AUCROC of 100%, specificity of 100%, sensitivity of 85.7%, an MCC equal to 0.925 and a Cohen’s kappa of 0.778. These results are encouraging enough to recognise deep learning algorithms as alternative avenues for detecting SARS CoV-2, as well as detecting regulatory motifs in the SARS CoV-2 genes. Co-author : Mr Charles Chimedza, School of Statistics and Actuarial Science, University of the Witwatersrand Charles.Chimedza@wits.ac.za