College of Science, Engineering and Technology

Deep learning for spatial multi-omics: predicting cardiomyocyte differentiation efficiency at single-cell resolution

2026-05-07T17:19:38Z

Deep learning for spatial multi-omics: predicting cardiomyocyte differentiation efficiency at single-cell resolution Kgabeng, Tumo Cardiovascular diseases remain the leading cause of global mortality, with limited regenerative capacity of adult cardiac tissue presenting significant therapeutic challenges. The primary cause of death worldwide is still cardiovascular diseases, and treating these conditions is extremely difficult due to the adult heart tissue's limited capacity for regeneration. Cardiomyocytes derived from human induced pluripotent stem cells (hiPSC CMs) present promising potential for cardiac regenerative medicine; however, existing differentiation protocols are highly inconsistent and do not have accurate predictive evaluation techniques. By integrating the analysis of temporal gene expression data and spatial transcriptomics, this study developed a novel hybrid deep learning architecture that combines Graph Neural Networks (GNNs) and Recurrent Neural Networks (RNNs) to predict the outcomes of cardiomyocyte differentiation. RNN components analysed temporal gene expression trajectories across 800 samples and 10 time points, while GNN components processed spatial transcriptomics data from 752 tissue spots to capture spatial relationships. Three fusion strategies - concatenation, attention-based, and ensemble approaches - were meticulously evaluated. With an accuracy of 96.67%, the ensemble fusion approach outperformed the state-of-the-art computational approaches by a significant margin (+13.47% compared to the top GNN approaches and +6.97% compared to specialised biological models). Keywords: Cardiomyocyte differentiation; Spatial transcriptomics, Spatial multi-omics; Single-cell biology; Deep learning; Graph Neural Networks; Recurrent Neural Networks; Stem cells; Artificial Intelligence; Cardiac biology

Estimating brittleness indexes from mechanical and petrographic characteristics of Norite

2026-03-27T13:20:00Z

Estimating brittleness indexes from mechanical and petrographic characteristics of Norite Molomo, Selaki Grace Norite is a coarse-grained plutonic rock that has been relatively understudied in terms of its mechanical and petrographic properties. This study investigates the brittleness of norite within the Eastern Limb of the Bushveld Igneous Complex (BIC), South Africa. However, there is a scarcity of studies that quantitatively link its petrographic characteristics to establish brittleness indices. The primary aim was to estimate brittleness indexes based on both mechanical and petrographic properties of norite, which is a significant rock type commonly found in the hanging walls of platinum mines. Given the recurring safety incidents, especially falls of ground and rock bursts in underground mining, understanding the brittleness of norite is essential for enhancing geotechnical designs and safety measures. Samples were collected from a 10-meter exposure along Mototolo Road in the Critical Zone of the Eastern Bushveld Complex, near the Anglo-American Platinum Mototolo Mine. Mechanical analysis involved laboratory testing, which includes uniaxial compressive strength (UCS), tensile strength, Young's modulus, and Poisson’s ratio, supported by numerical simulations and multivariate regression models. The results indicate that norite exhibits high compressive strength and low ductility, with brittleness indexes effectively predicted using combinations of strength parameters. Mineralogical investigations were done using thin-section petrography to evaluate grain texture, contact nature, and mineral composition. It was observed that coarse and medium grain textures significantly influence brittleness, whereas grain contact type alone lacks predictive power. The main contribution of this work is the development of integrated predictive models that use both mechanical and mineralogical data. While the use of surface samples presents a limitation, their geological equivalence to underground norite supports the relevance of the findings for subsurface application. The findings enhance the understanding of the structural performance of norite and suggest practical recommendations for underground mine design. This research further contributes to improved and safer mining operations.

Investigating how Artificial Intelligence (AI) can be leveraged to optimise the distribution and allocation of social grants in South Africa

2026-04-07T09:36:54Z

Investigating how Artificial Intelligence (AI) can be leveraged to optimise the distribution and allocation of social grants in South Africa Hlatshwayo, Mthokozisi Alfred This study investigates the potential of Artificial Intelligence (AI) technologies to enhance the distribution and allocation of social grants in South Africa, addressing inefficiencies, high costs, and administrative complexities. Social grants are vital for reducing poverty and improving welfare, but the current system faces challenges such as lengthy processing times, inconsistent data, and limited accessibility. A qualitative research design was employed, utilising semi-structured interviews with 20 stakeholders, including policymakers, officials from the South African Social Security Agency (SASSA), and beneficiaries. Data analysis used thematic analysis to assess the feasibility of AI integration. Pilot studies of Electronic Know Your Client (EKYC) and biometric verification across select provinces provided empirical insights. In order to enhance decision-making, lower fraud, expedite beneficiary verification, and optimise resource allocation within South Africa's social grant ecosystem, this study explores the potential applications of artificial intelligence (AI) technologies, including machine learning, predictive analytics, natural language processing, automated document verification, and anomaly detection. The study looks at international best practices, assesses the possible advantages and disadvantages of integrating AI in South Africa, and reviews policy preparedness in light of the nation's larger Fourth Industrial Revolution (4IR) goal. Findings reveal that AI technologies could mitigate inefficiencies by automating data verification, enhancing fraud detection, and streamlining decision-making processes. Pilot implementations of Electronic Know Your Client (EKYC) and biometric systems across Gauteng, Eastern Cape, and North West provinces demonstrated varied outcomes. Gauteng saw a 5% improvement in fraud detection but faced challenges with system integration. The Eastern Cape struggled with downtimes, while the North West achieved a 10% reduction in identity theft but encountered transaction inaccuracies. This research contributes a comprehensive AI integration model tailored to South Africa’s socio-economic context, providing actionable recommendations for policymakers. Key considerations include robust infrastructure, workforce training, regulatory updates, and ethical safeguards to ensure fair and inclusive AI deployment. The study advances the discourse on AI in public administration, offering practical insights for achieving greater efficiency, transparency, and equity in social grant distribution.; Lolu cwaningo luhlola ukuhlanganiswa kobuchwepheshe be-Artificial Intelligence (AI) ohlelweni lokusabalalisa izibonelelo zenhlalakahle eNingizimu Afrika ukuze kuthuthukiswe ukusebenza kahle, kuncishiswe ukukopela, futhi kwandiswe ukufinyeleleka. Uhlelo lwezibonelelo zenhlalakahle lwaseNingizimu Afrika, olulawulwa yi-South African Social Security Agency (SASSA), lubalulekile ekunciphiseni ububha nasekuthuthukiseni inhlalakahle yezenhlalo. Nokho, izinselelo ezinjengokungasebenzi kahle kokuphatha, ukukopela, nokulinganiselwa kwezinsiza kuthinta ukusebenza kahle kwalo hlelo. Lolu cwaningo lusebenzisa indlela yokucwaninga yekhwalithethivu, lusebenzisa izingxoxo ezihlelekile neziphathimandla ezibalulekile, okuhlanganisa izikhulu ze-SASSA, abenzi benqubomgomo, nabathola izibonelelo. Ukuhlaziywa kwezindikimba zedatha, okuhlangene nocwaningo lwamapayona lwe-Electronic Know Your Client (EKYC) kanye nebiometric verification ezifundazweni ezikhethiwe, kuhlaziywe imiphumela yocwaningo. Ucwaningo lwapayona lubonise amandla e-AI ekutholeni ukukopela, kwathi eGauteng kwaba nokwehla ngo-5% kokukopela, kwathi eNyakatho Ntshonalanga kwaba nehlazo lokuncipha kuka-10% ekuntshontshweni kobunikazi. Nokho, izinselelo ezifana nokwehla kwezinhlelo zesistimu nokungahlangani kahle kwezinhlelo eMpumalanga Kapa ziveza isidingo sokuthuthukiswa kwezingqalasizinda. Imiphumela ikhombisa ukuthi i-AI ingathuthukisa kakhulu ukuphathwa kwezibonelelo zenhlalakahle ngokuzenzekelayo ekuqinisekisweni kwabamukeli, ekunciphiseni izikhathi zokucubungula, nasekukholeni ukusebenza kwamasistimu atholayo. Nokho, ukuthuthukiswa kwe-AI okuphumelelayo kudinga ingqalasizinda eqinile, ukuqeqeshwa kwabasebenzi, kanye nokuqapha kwezomthetho ukuze kuqinisekiswe ukusetshenziswa okuzuzisayo nokulungile. Lolu cwaningo lunikeza isibonelo sokuqaliswa kwe-AI esenzelwe isimo seNingizimu Afrika, lunikeza izincomo ezisebenzayo kubenzi benqubomgomo. Ucwaningo lufaka isandla engxoxweni enkulu mayelana nokwethulwa kwe-AI ezinsizeni zomphakathi, lugqamisa isidingo sendlela elinganiselayo eqhathanisa izinzuzo zokusebenza kahle kanye nemithelela yesimilo nezenhlalo. Ngokubhekana nezinselelo zokwethulwa kwe-AI, lolu cwaningo lunikeza isiqondiso sendlela yokusebenzisa i-AI ukwenza uhlelo lwezibonelelo zenhlalakahle eNingizimu Afrika lube olusebenzayo, olungcono, nolungagwegwesi.; Patlisiso ena e hlahloba ho kenyelletsoa ha mahlale a Artificial Intelligence (AI) tsamaisong ea kabo ea meputso ea sechaba Afrika Boroa ho ntlafatsa ts'ebetso, ho fokotsa bomenemene, le ho ntlafatsa phihlello. Tsamaiso ea meputso ea sechaba ea Afrika Boroa, e tsamaisoang ke South African Social Security Agency (SASSA), e bohlokoa ho fokotseng bofuma le ho ntlafatsa boiketlo ba sechaba. Leha ho le joalo, mathata a kang ho se sebetse hantle tsamaisong, bomenemene, le mefokolo ea lisebelisoa a sitisa katleho ea tsamaiso ena. Patlisiso ena e sebelisa mokhoa oa lipatlisiso tsa boleng bo phahameng, e sebelisa lipuisano tse hlophisitsoeng le ba boholong, ho kenyeletsoa liofisiri tsa South African Social Security Agency (SASSA), ba etsang melao, le ba amohelang meputso ea sechaba. Tlhatlhobo ea data e entsoeng ka ho sekaseka lihlooho tsa bohlokoa, hammoho le teko ea pele ea Electronic Know Your Client (EKYC) le biometric verification liprofinseng tse ling, e fana ka leseli ka bokhoni ba AI tsamaisong ena. Teko ena e bontšitse hore AI e ka fokotsa bomenemene, moo Gauteng ho bileng le phokotso ea 5% ea bomenemene, ha Profense ya North West e bone phokotso ea 10% bosholu ba boitsebiso. Leha ho le joalo, mathata a kang ho putlama ha tsamaiso le mathata a kopano ea sistimi Profinseng ea Eastern Cape a bontša tlhoko ea lintlafatso tsa motheo. Liphuputso li bontša hore AI e ka ntlafatsa tsamaiso ea meputso ea sechaba ka ho nolofatsa ts'ebetso ea netefatso ea ba amohelang meputso, ho fokotsa nako ea ts'ebetso, le ho ntlafatsa lits'ebetso tsa tlhahlobo ea bomenemene. Leha ho le joalo, katleho ea AI e hloka hore ho be le motheo o tiileng oa meaho, koetliso ea basebetsi, le taolo e matla ho netefatsa hore AI e sebelisoa ka toka le ka nepo. Patlisiso ena e fana ka mohlala oa ho kenya AI ts'ebetsong ka mokhoa o lumellanang le maemo a Afrika Boroa, e fana ka likhothaletso tse sebetsang ho ba etsang melao le ba ikarabellang ho tsamaiso ea meputso ea sechaba. Patlisiso ena e kenya letsoho lipuisanong tse pharalletseng mabapi le AI le lits'ebeletso tsa sechaba, e hatisa tlhokahalo ea mokhoa o leka-lekaneng o hlahlobang melemo ea ts'ebetso hammoho le litlamorao tsa boitšoaro le tsa sechaba. Ka ho sebetsana le mathata a kenyelletsong ea AI, patlisiso ena e fana ka tataiso ea ho sebelisa AI ka tsela e ntlafatsang le e lumellanang le litlhoko tsa sechaba tsa Afrika Boroa. Abstract in English, Isizulu and Sesotho

Compartmental model for the spread of infectious disease with hererogenous population: a case study of COVID-19 and Lassa fever

2026-03-20T09:07:39Z

Compartmental model for the spread of infectious disease with hererogenous population: a case study of COVID-19 and Lassa fever Oluwagunwa, Abiodun Peter This study investigates the transmission dynamics of COVID-19 and Lassa fever, with particular attention to the risks and implications of co-infection. By dividing the human population into epidemiological compartments, the models capture the real course of disease spread in a structured and realistic way. The mathematical correctness of the models was validated by proving positivity, boundedness, and reliability of solutions for public health interpretation. For COVID-19, the basic SYR framework was analysed to obtain the reproduction number RY , and the model was further extended to an SEAIHR structure to include exposed, asymptomatic, infectious, and hospitalised individuals. For Lassa fever, a deterministic compartmental model was developed and subjected to stability analysis. Numerical simulations were carried out for both diseases to assess intervention strategies and transmission behaviour. The COVID–19 analysis revealed that the disease can be eliminated when RY < 1, but once RY > 1, infection becomes persistent. The extended SEAIHR model also produced the overall reproduction number R0, showing how reductions in contact rates, timely detection, effective hospital care, and faster recovery can substantially suppress transmission. For Lassa fever, the disease-free equilibrium remained stable only when the reproduction number was kept below one. Simulations highlighted the significant influence of asymptomatic carriers and showed that no single intervention— whether treatment, health education, or rodent control—can fully control the disease on its own. Instead, the most meaningful reduction in cases occurred when human-focused measures were combined with strong rodent control, leading to the elimination of infectious rodents by the 35th day. Together, these results emphasise the urgency of studying co-infections in regions such as West Africa, where both diseases circulate at the same time. Incorporating optimal control theory provides a systematic and cost-effective framework for coordinating interventions across multiple pathways of transmission. This study therefore deepens our understanding of the dynamics of both COVID– 19 and Lassa fever and offers practical insights for improving interventions, epivdemic preparedness, and public health responses. The centre-manifold analysis further shows that both models experience a forward transcritical bifurcation at R0 = 1: once transmission exceeds this threshold, a stable endemic state emerges. This reinforces a critical message—maintaining transmission below the threshold is essential for preventing long-term persistence of either disease.