Non-vector data and SVM? My research is on antimicrobial peptide classification and prediction. I have gathered peptide sequences of lengths ranging from 10 - 200 and classified them using different machine learning algorithms. The algorithm with the highest performance in terms of ACC, SEN, SPC and MCC was the SVM. My negative and positive datasets are balanced, each with about 500 samples. The total number of attributes was 126. I want to know if my dataset can be classified as a non-vector dataset? Also, if the reason SVM had the best performance is that kernel methods are more appropriate when dealing with non-vector inputs?
P.S. I major in biotechnology and my knowledge on machine learning is very general and basic.
 A: If you were able to easily run SVM and other ML models then your dataset is not "non-vector like". You had to somehow represent your data in a way that is understandable by SVM and other methods - if you did encode them to vector format then your data is vector like. If you, however used some tricky kernel, which fitted model to raw, non vector data (such as text, or graph) you could advocate the term "non-vector like", however this is purely linguistic as "under the hood", SVM always works on vectors, sometimes in a very weird spaces, yet still vector, Hilbert spaces.
A: (I just saw this, so if this response is too late, I am sorry.)
Yes, your data is non-vector data. 
Since you are able to compute similarities between instances you can cluster them. Once you can cluster them, you can do a number of things (from simplest to more complex): 
-Cluster and use centroids for classification: e.g. http://ieeexplore.ieee.org/document/4401090/
-Use RBF Networks where the RBF units use your similarity function instead of the regular Gaussians.
-Use SVMs with your similarity measure in the kernels. 
...
Hope this helps. 
