Deep learning for low-data drug discovery: Hurdles and opportunities.

Deep learning is becoming increasingly relevant in drug discovery, from de novo design to protein structure prediction and synthesis planning. However, it is often challenged by the small data regimes typical of certain drug discovery tasks. In such scenarios, deep learning approaches-which are notoriously 'data-hungry'-might fail to live up to their promise. Developing novel approaches to leverage the power of deep learning in low-data scenarios is sparking great attention, and future developments are expected to propel the field further. This mini-review provides an overview of recent low-data-learning approaches in drug discovery, analyzing their hurdles and advantages. Finally, we venture to provide a forecast of future research directions in low-data learning for drug discovery.

Direct Microscopic Proof of the Fermi Level Pinning Gas-Sensing Mechanism: The Case of Platinum-Loaded WO3.

It is widely known that the sensing characteristics of metal oxides are drastically changed through noble metal oxide surface additives. Using operando infrared spectroscopy it was identified that the Fermi level pinning mechanism dominates the sensor response of platinum-loaded WO3. Spectroscopy, however, provides information about the sample only on average. Traditional microscopy offers structural information but is typically done in vacuum and on unheated sensors, very different than the operation conditions of metal oxide gas sensors. Here, state-of-the-art in situ scanning transmission electron microscopy offers spatially resolved information on heated samples at atmospheric pressure in varying gas atmospheres. As a result it was possible to directly couple microscopically observed structural changes in the surface noble metal nanoclusters with IR spectra and sensor responses. On the basis of the findings, the dominant Fermi level pinning mechanism could be validated. The presented work demonstrates the benefits of coupling in situ microscopy with operando spectroscopy in order to elucidate the sensing mechanism of metal oxides.

Rhodium Oxide Surface-Loaded Gas Sensors.

In order to increase their stability and tune-sensing characteristics, metal oxides are often surface-loaded with noble metals. Although a great deal of empirical work shows that surface-loading with noble metals drastically changes sensing characteristics, little information exists on the mechanism. Here, a systematic study of sensors based on rhodium-loaded WO3, SnO2, and In2O3-examined using X-ray diffraction, high-resolution scanning transmission electron microscopy, direct current (DC) resistance measurements, operando diffuse reflectance infrared Fourier transform (DRIFT) spectroscopy, and operando X-ray absorption spectroscopy-is presented. Under normal sensing conditions, the rhodium clusters were oxidized. Significant evidence is provided that, in this case, the sensing is dominated by a Fermi-level pinning mechanism, i.e., the reaction with the target gas takes place on the noble-metal cluster, changing its oxidation state. As a result, the heterojunction between the oxidized rhodium clusters and the base metal oxide was altered and a change in the resistance was detected. Through measurements done in low-oxygen background, it was possible to induce a mechanism switch by reducing the clusters to their metallic state. At this point, there was a significant drop in the overall resistance, and the reaction between the target gas and the base material was again visible. For decades, noble metal loading was used to change the characteristics of metal-oxide-based sensors. The study presented here is an attempt to clarify the mechanism responsible for the change. Generalities are shown between the sensing mechanisms of different supporting materials loaded with rhodium, and sample-specific aspects that must be considered are identified.