Table 1.
All reviewed frameworks, backends, programing languages, and types of processing units are summarised.
| Peak Alignment/Identification | Compound and Structure Identification/Quantification | Data Analysis/Omics Integration | Total | ||
|---|---|---|---|---|---|
| Framework** | Keras | 2 | 5 | 3 | 10 |
| MXNet | 0 | 0 | 1 | 1 | |
| H2O* | 0 | 0 | 2 | 2 | |
| FANN/RPROP | 0 | 1 | 1 | 2 | |
| DLT | 1 | 2 | 0 | 3 | |
| Backend | TensorFlow | 3 | 7 | 2 | 12 |
| Theano | 0 | 1 | 1 | 2 | |
| PyTorch | 1 | 0 | 0 | 1 | |
| MXNet | 0 | 0 | 3 | 3 | |
| Others | 1 | 3 | 3 | 7 | |
| Programming Language | Python | 4 | 8 | 3 | 15 |
| R | 0 | 0 | 2 | 2 | |
| MATLAB | 1 | 2 | 0 | 3 | |
| C | 0 | 1 | 1 | 2 |
DLT = Deep Learning Toolbox in MATLAB, note that this also includes old implementation Neural Network Toolbox
Others = MATLAB toolboxes, RPROP
* Since 2018, H2O no longer uses MXNet or TensorFlow as backend. As these studies were conducted prior to 2018, we assumed (according to the source code) that the framework still employed default backend, which is Apache MXNet)
** Some studies employed TensorFlow directly as their framework in Python. Therefore, the number of studies in framework rows are not matched with number of Backend rows.