Contents I. Introduction
Ion channels are a group of proteins in a cell that can open and close based on stimuli response [1], [2]. When these ion channels open, the electric signal passes through them which act as a response to a particular action. The signal value changes very frequently in order of 0.1 milliseconds. Every ion channel has a different function, thus opening and closing of these channels is dependent on the stimulus-response to a particular action. For example, different ion channels would be open in case someone witnesses a horrific event in comparison to when someone is playing with their pet. Many studies have worked upon the different aspects of the ion channels. Predicting the type of ion channel is one such study [3], [4]. The authors of [3] propose a support vector machine (SVM) based classification of ion channels into four classes of protein wiz. potassium, sodium, calcium, and anion. The study makes use of Analysis Of Variance (ANOVA) for important feature selection. The study also proposes a model that can differentiate an ion channel from a non-ion channel with an accuracy of 86.6%. The model can classify the ion channels into the four given classes with an accuracy of 87.8% and further claims that it could be used to develop target-based drugs. Whereas [4] is an extension to this study, as it uses residue sequence information and proposes a novel feature selection method using the physiological correlation among the residues. All these studies discuss how machine learning could help identify the open channels and then that analysis could further be used to develop target-based drugs i.e. using the machine learning for target-based drugs passively. But [5], gave a direct application of machine learning for target-based drugs by using an SVM to predict a possible drug target using the ion channels. The study came up with a conclusion that 4 out of 5 most occurring ion channel proteins were related to nerve diseases, thus indicating a potential to develop drugs for nerve diseases. Also, [6] gave a direct application by predicting the ion channel genes and their types using Random Forests classifier. Moreover, the chemicals and genetics could play a role in disrupting the functioning of the ion channels which further could harm the organisms [7]. A work was done by the authors of [8] to establish a relationship between ion channels and axonal conduction velocity by capturing the activity of cortical neurons propagating along the axons, using a device that uses photo-lithography. Another study used the Binary Kernel Discrimination approach to rank the set of ion channel compounds as per the probability of their opening during activity [9]. Both structural and qualitative aspects of the compound [10] are used to classify the ion channels. Whereas, [11] provided a detailed review of the ion channel-based electrochemical techniques. Apart from this, [12] published a ground-breaking work which uses a deep learning model based on convolutional neural network and long-short term memory to detect single-molecule events from the data collected using the patch-clamp electrophysiology. This work was able to capture the movements of a single ion in real-time. This paper extends the same methodology to predict the number of open ion channels at a given point of time in a cell. The major contributions of this paper include:
Providing a benchmark analysis to classify the ion-channels by using different gradient boosting frameworks. This also indicates the benefit of using gradient boosting techniques against a primitive classification algorithm like KNN.
The robustness for the real-world scenarios is introduced by feature engineering that handles the real-world noisy data.
The proposed study can classify ion channel proteins into some classes based on open channels at aparticular time, opening research scope for biologists to develop target-based drugs.