Tom Lee (Research Fellow, MD Group, SCMB)

Title: The molecular origin of anisotropic emission in an organic light-emitting diode

Organic light-emitting diodes (OLEDs) are increasingly used in electronic displays and related technologies. A major factor in the efficiency of these devices is the loss of photons to reabsorption within the light-emitting layer. These losses can be minimized by increasing the degree of horizontal alignment of the transition dipole moment vectors (TDVs) associated with the phosphorescent emitter molecules. Here, non-equilibrium molecular dynamics simulations have been used to model the induction of anisotropy in the TDV orientations within an OLED emission layer formed by vapor deposition. Two emitter species were compared: fac-tris(2-phenylpyridine)iridium(III) (Ir(ppy)3), which experimentally shows isotropic emission, and bis(2-phenylpyridine)(acetylacetonate)iridium(III) (Ir(ppy)2(acac)), which displays increased efficiency due to apparent alignment of the TDV. It is found that during deposition, the molecular axis of both emitters becomes preferentially aligned with the normal to the surface. The simulations suggest that the experimentally observed differences in the efficiency of these systems is due to the fact that the three TDVs of Ir(ppy)3 are nearly orthogonal, canceling the effect of the molecular alignment, whereas the two TDVs of Ir(ppy)2(acac) both preferentially lie parallel to the surface.

Kanupriya Tiwari (PhD, Nielsen Group, AIBN)

Title: Bayesian Neural Networks for predicting alternative gene splicing patterns. 

Alternative Splicing significantly increases the coding potential of our genetic material by generating several mature transcripts from a single pre-mRNA.  More than 95% of human genes are estimated to be alternatively spliced regulating a wide spectrum of physiologically important events such as cell cycle control, apoptosis, stem cell pluripotency and lineage specification, regulation of epithelial to mesenchymal transition, generation of neuronal plasticity and response to heat shock stress. Dysregulation of splicing has been observed to be involved in the pathology of several diseases such as thalassaemia, myelodysplasia, spinal muscular atrophy, FTDP-17, Alzheimer’s disease and also several forms of cancer. The combinatoric action of DNA sequence dependent features along with trans factor expression and epi-genetic features has complicated the process of in silico modelling of splicing which would be useful for prediction of splice patterns for previously uncharacterized events as well generating hypotheses about the effect of changing regulatory factors such as splice site mutations or splice factor knockouts for example.  Fortunately, machine learning methods lend themselves well to such problems of deriving predictive models when the underlying mechanisms are insufficiently defined or too complex to be represented as meaningful mathematical relationships. Neural networks are a powerful, biologically inspired method to learn predictive models from observational data. However, the use of conventional feed forward networks that provide point estimates of parameters and predictions does not give us an insight about the confidence in model predictions, i.e. in cases where the model is presented with a test case that lies far away from the distribution of the training data, we can never be sure whether the model is making an educated guess or just a random prediction. In such cases, the ideal behaviour would be that the model provide us an estimate of the degree of confidence in its prediction. This problem can be tackled by the use of Bayesian modelling to account for the uncertainty in the network parameters , i.e., a Bayesian Neural Network framework. In this presentation, I will talk about using a Bayesian neural network based model for predicting gene splicing patterns using sequence and splicing protein expression patterns.