Identifying diverse metal oxide nanomaterials with lethal effects on embryonic zebrafish using machine learning

• Richard Liam Marchese Robinson,
• Haralambos Sarimveis,
• Philip Doganis,
• Xiaodong Jia,
• Marianna Kotzabasaki,
• Christiana Gousiadou,
• Stacey Lynn Harper and
• Terry Wilkins

Beilstein J. Nanotechnol. 2021, 12, 1297–1325, doi:10.3762/bjnano.12.97

• -fertilisation data were found to be harder to predict, which could reflect different exposure routes. Hence, subsequent analysis focused on the prediction of excess lethality at 120 hours-post fertilisation. The use of two data augmentation approaches, applied for the first time in nano-QSAR research, was

• these modelling results are on truly external data, which were not used to select the single descriptor model. This will require further laboratory work to generate comparable data to those studied herein. Keywords: data augmentation; embryonic zebrafish; machine learning; nanosafety; nano-QSAR

• our study is that we explored two data augmentation approaches which, to the best of our knowledge, have never previously been applied in published nano-QSAR research, as a means of addressing the widely known issues with limited availability of suitable data for nano-QSAR development [20][33][34

The role of convolutional neural networks in scanning probe microscopy: a review

• Ido Azuri,
• Irit Rosenhek-Goldian,
• Neta Regev-Rudzki,
• Georg Fantner and
• Sidney R. Cohen

Beilstein J. Nanotechnol. 2021, 12, 878–901, doi:10.3762/bjnano.12.66

• over-fitting. Since there are more examples in the data set, a more general model is obtained. Collecting more data would increase the data set size, but this may not be feasible. Another solution is data augmentation. Data augmentation [9][10][13] is a method to increase the sample data set size, for

• approaches to improve them, for example by enhancing contrast and increasing the amount of training data by data-augmentation methods mentioned above [7][97], or by using classical image reconstruction methods such as nearest neighbor, bilinear, or bicubic interpolations [99]. One way to exploit the power of