Neural Information Processing - Letters and Reviews Vol. 3, No. 2, May 2004 31 Transductive Support Vector Machines and Applications in Bioinformatics for Promoter Recognition Nikola Kasabov and Shaoning Pang Knowledge Engineering & Discover Research Institute Auckland University of Technology, Private Bag 92006, Auckland 1020, New Zealand E-mail: nkasabov@aut.ac.nz, spang@aut.ac.nz (Submitted on March 24, 2004) Abstract— This paper introduces a novel Transductive Support Vector Machine (TSVM) model and compares it with the traditional inductive SVM on a key problem in Bioinformatics - promoter recognition. While inductive reasoning is concerned with the development of a model (a function) to approximate data from the whole problem space (induction), and consecutively using this model to predict output values for a new input vector (deduction), in the transductive inference systems a model is developed for every new input vector based on some closest to the new vector data from an existing database and this model is used to predict only the output for this vector. The TSVM outperforms by far the inductive SVM models applied on the same problems. Analysis is given on the advantages and disadvantages of the TSVM. Hybrid TSVM-evolving connections systems are discussed as directions for future research. Keywords—Transductive SVM, Inductive SVM, Promoter recognition, Motif, Promoter Vacabulary 1. Inductive and Transductive Inferences Most of the learning models and systems in artificial intelligence apply inductive inference where a model (a function) is derived from data and this model is further applied on new data. [1]. This is the case in the area of soft computing, [2] [3] [4-7], and particularly - in neuro-fuzzy reasoning systems [8, 9] [10], and in support vector machines (SVM) [11]and in their numerous applications (see for example [12]). The model is created without taking into account any information about a particular new data vector. The new data would fit into the model to certain degree (an error is estimated). The model is in most cases a global model, covering the whole problem space. Creating a global model (function) that would be valid for the whole problem space is a difficult task and in most cases - it is not necessary. In some local learning systems (see for example [13] [14]) that include the evolving connectionist systems (ECOS) [15] the global model consists of many local models (rules) that collectively cover the whole space and are adjusted individually on new data. The output for a new vector is calculated based on the activation of one or several neighboring local models (rules). The inductive learning and inference approach is useful when a global model ("the big picture") of the problem is needed even in its very approximate form, when incremental, on-line learning is applied to adjust this model on new data and trace its evolution. Generally speaking, inductive inference is concerned with the estimation of a function (a model) based on data from the whole problem space and using this model to predict output values for a new input vector, which can be any point in this space (deduction) - Fig.1. Most of the statistical, connectionist and fuzzy learning methods, such as SVM, MLP; RBF; ANFIS (see [10]; [16]) and ECOS [15] that include DENFIS [17], EFuNN [18, 19] and many more, have been developed and tested on inductive reasoning problems. In contrast to the inductive inference, transductive inference methods estimate the value of a potential model (function) only for a single point of the space (the new data vector) utilizing additional information related to this point [11]. This approach seems to be more appropriate for clinical and medical applications of learning systems, where the focus is not on the model, but on the individual patient data. And it is not so important what LETTER