Few-Shot Learning for Low-Data Drug Discovery.

The discovery of new hits through ligand-based virtual screening in drug discovery is essentially a low-data problem, as data acquisition is both difficult and expensive. The requirement for large amounts of training data hinders the application of conventional machine learning techniques to this problem domain. This work explores few-shot machine learning for hit discovery and lead optimization. We build on the state-of-the-art and introduce two new metric-based meta-learning techniques, Prototypical and Relation Networks, to this problem domain. We also explore using different embeddings, namely, extended-connectivity fingerprints (ECFP) and embeddings generated through graph convolutional networks (GCN), as inputs to neural networks for classification. This study shows that learned embeddings through GCNs consistently perform better than extended-connectivity fingerprints for toxicity and LBVS experiments. We conclude that the effectiveness of few-shot learning is highly dependent on the nature of the data. Few-shot learning models struggle to perform consistently on MUV and DUD-E data, in which the active compounds are structurally distinct. However, on Tox21 data, the few-shot models perform well, and we find that Prototypical Networks outperform the state-of-the-art, which is based on the Matching Networks architecture. Additionally, training these networks is substantially faster (up to 190%) and therefore takes a fraction of the time to train for comparable, or better, results.

The effect of alloying elements on the properties of pressed and non-pressed biodegradable Fe-Mn-Ag powder metallurgy alloys.

Current trends in the biodegradable scaffold industry call for powder metallurgy methods in which compression cannot be applied due to the nature of the scaffold template itself and the need to retain the shape of an underlying template throughout the fabrication process. Iron alloys have been shown to be good candidates for biomedical applications where load support is required. Fe-Mn alloys were researched extensively for this purpose. Current research shows that all metallurgical characterisation and corrosion test on Fe-Mn and Fe-Mn-Ag non pre-alloyed powder alloys are performed on alloys which are initially pressed into greens and subsequently sintered. In order to combine the cutting-edge field of biodegradable metallic alloys with scaffold production, metallurgical characterisation of pressed and non-pressed Fe, Fe-Mn and Fe-Mn-Ag sintered elemental powder compacts was carried out in this study. This was performed along with determination of the corrosion rate of the same alloys in in vitro mimicking solutions. These solutions were synthesised to mimic the osteo environment in which the final scaffolds are to be used. Both pressed and non-pressed alloys formed an austenite phase under the right sintering conditions. The corrosion rate of the non-pressed alloy was greater than that of its pressed counterpart. In a potentiodynamic testing scenario, addition of silver to the alloy formed a separate silver phase which galvanically increased the corrosion rate of the pressed alloy. This result wasn't replicated in the non-pressed alloys in which the corrosion rate was seen to remain similar to the non-silver-bearing alloy counterparts.

Influence of cold rolling on in vitro cytotoxicity and electrochemical behaviour of an Fe-Mn-C biodegradable alloy in physiological solutions.

The properties of cold-worked Fe-13Mn-1.2C steel, as candidate material for scaffolding and stenting applications, have been investigated. The study of the electrochemical corrosion susceptibility of Fe-13Mn-1.2C alloy in protein bearing and non-protein bearing physiological solutions, revealed that there were no differences between the as-received, 10% and 20% cold worked Fe-13Mn-1.2C samples. Although protein addition reduces the overall corrosion rate in static immersion degradation tests for both the cold worked and non-cold worked alloys, there were no discernible differences in the corrosion rates of samples with different percentages of cold work deformations. Similarly, potentiodynamic testing showed no differences between the corrosion rates in solutions with and without protein addition. Atomic absorption spectroscopy (AAS) results-post static immersion-showed similar values of Fe and Mn concentrations in the electrolyte for all the investigated conditions. Cold working was found to increase Grain Average Misorientation (GAM) and deformation twins within the steel, and, consequently, this led to an increase in the elastic modulus. Hence, cold-rolling may be used to achieve smaller sections (volumes) in order to support the equivalent load of the non-cold worked counterpart, giving a larger surface area to the volume ratio, thereby increasing the corrosion rate, and, in turn, rendering the degradation process shorter. When considering cytocompatibility in vitro, the collected supernatant particulate free Fe-13Mn-1.2C steel electrolytes were seen to be equally cytocompatible with no differences being observed between the different percentage cold work conditions. The presence of solid 80 µm size particles in the seeded elutions were seen to change the results and render the Fe-13Mn-1.2C steel non-cytocompatible.

Biocalcifying Bacillus subtilis cells effectively consolidate deteriorated Globigerina limestone.

Microbially induced calcite precipitation occurs naturally on ancient limestone surfaces in Maltese hypogea. We exploited this phenomenon and treated deteriorated limestone with biocalcifying bacteria. The limestone was subjected to various mechanical and physical tests to present a statistically robust data set to prove that treatment was indeed effective. Bacillus subtilis conferred uniform bioconsolidation to a depth of 30 mm. Drilling resistance values were similar to those obtained for freshly quarried limestone (9 N) and increased up to 15 N. Treatment resulted in a high resistance to salt deterioration and a slow rate of water absorption. The overall percentage porosity of treated limestone varied by ±6 %, thus the pore network was preserved. We report an eco-friendly treatment that closely resembles the mineral composition of limestone and that penetrates into the porous structure without affecting the limestones' natural properties. The treatment is of industrial relevance since it compares well with stone consolidants available commercially.