Biomedical informatics includes all techniques regarding the development of data analytics, mathematical modeling, and computational simulation for the study of biological systems. In recent years, we've witnessed huge leaps in biological computing that has resulted in large, information-rich resources being at our disposal. These cover domains such as anatomy, modeling (3D printers), genomics, and pharmacology, among others.

One of the most famous success stories of biomedical informatics is from the domain of genomics. The Human Genome Project (HGP) was an international research project with the objective of determining the full sequence of human DNA. This project has been one of the most important landmarks in computational biology and has been used as a base for other projects, including the Human Brain Project, which is determined to...

Genomics data covers all data related to DNA on living things. Although in this thesis we will also use other types of data like transcriptomic data (RNA and miRNA), for convenience purposes, all data will be termed as genomics data. Research on human genetics found a huge breakthrough in recent years due to the success of the HGP (1984-2000) on sequencing the full sequence of human DNA.

One of the areas that have been helped a lot due to this is the research of all diseases related to genetics, including cancer. Due to various biomedical analyses done on DNA, there exist various types of -omics or genomics data. Here are some types of -omics data that were crucial to cancer analysis:

• Raw sequencing data: This corresponds to the DNA coding of whole chromosomes. In general, every human has 24 types of chromosomes in each cell of their body,...

In this section, we will discuss how to configure DL4J, ND4s, Spark, and ND4J before getting started with the coding. The following are prerequisites when working with DL4J:

• Java 1.8+ (64-bit only)
• Apache Maven for automated build and dependency manager
• IntelliJ IDEA or Eclipse IDE
• Git for version control and CI/CD

The following libraries can be integrated with DJ4J to enhance your JVM experience while developing your ML applications:

• DL4J: The core neural network framework, which comes up with many DL architectures and underlying functionalities.
• ND4J: Can be considered as the NumPy of the JVM. It comes with some basic operations of linear algebra. Examples are matrix creation, addition, and multiplication.
• DataVec: This library enables ETL operation while performing feature engineering.
• JavaCPP: This library acts as the bridge between Java...

In the previous section, we have seen what our data (that is, features and labels) looks like. Now in this section, we try to classify those samples according to their labels. However, as we have seen, DL4J needs the data in a well-defined format so that it can be used to train the model. So let us perform the necessary preprocessing and feature engineering.

Since we do not have any unlabeled data, I would like to select some samples randomly for test. Well, one more thing is that features and labels come in two separate files. Therefore, we can perform the necessary preprocessing and then join them together so that our pre-processed data will have...

Now that we have solved the Titanic survival prediction problem with an acceptable level of accuracy, there are other practical aspects of this problem and overall deep learning phenomena that need to be considered too. In this section, we will see some frequently asked questions that might be already in your mind. Answers to these questions can be found in Appendix A.

1. Can't we use MLP to solve the cancer type prediction by handling this too high-dimensional data?
2. Which activation and loss function can be used with RNN type nets?
3. What is the best way of recurrent net weight initialization?
4. Which updater and optimization algorithm should be used?
5. In the Titanic survival prediction problem, we did not experience good accuracy. What could be possible reasons and how can we improve the accuracy?
6. The predictive accuracy for cancer type prediction...

In this chapter, we saw how to classify cancer patients on the basis of tumor types from a very-high-dimensional gene expression dataset curated from TCGA. Our LSTM architecture managed to achieve 100% accuracy, which is outstanding. Nevertheless, we discussed many aspects of DL4J, which will be helpful in upcoming chapters. Finally, we saw answers to some frequent questions related to this project, LSTM network, and DL4J hyperparameters/nets tuning.

In the next chapter, we will see how to develop an end-to-end project for handling a multilabel (each entity can belong to multiple classes) image classification problem using CNN based on Scala and the DL4J framework on real Yelp image datasets. We will also discuss some theoretical aspects of CNNs before getting started. Nevertheless, we will discuss how to tune hyperparameters for better classification results.

Answer to question 1: The answer is yes, but not very comfortably. That means a very deep feedforward network such as deep MLP or DBN can classify them with too many iterations.

However, also to speak frankly, MLP is the weakest deep architecture and is not ideal for very high dimensions like this. Moreover, DL4J has deprecated DBN since the DL4J 1.0.0-alpha release. Finally, I would still like to show an MLP network config just in case you want to try it:

// Create network configuration and conduct network training
MultiLayerConfiguration MLPconf = new NeuralNetConfiguration.Builder().seed(seed)
                .optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
                .updater(new Adam(0.001)).weightInit(WeightInit.XAVIER).list()
                .layer(0,new DenseLayer.Builder().nIn(numInputs).nOut(32)
                        .weightInit...