Multi-objective Optimisation of Cancer Chemotherapy Using Evolutionary Algorithms Andrei Petrovski and John McCall School of Computer and Mathematical Sciences, The Robert Gordon University, St. Andrew Street, ABERDEEN, AB25 1HG, United Kingdom {ap,jm}@scms.rgu.ac.uk Abstract. The main objectives of cancer treatment in general, and of cancer chemotherapy in particular, are to eradicate the tumour and to prolong the patient survival time. Traditionally, treatments are optimised with only one objective in mind. As a result of this, a particular patient may be treated in the wrong way if the decision about the most appropriate treatment objective was inadequate. To partially alleviate this problem, we show in this paper how the multi-objective approach to chemotherapy optimisation can be used. This approach provides the oncologist with versatile treatment strategies that can be applied in ambiguous cases. However, the conflicting nature of treatment objectives and the non-linearity of some of the constraints imposed on treatment schedules make it difficult to utilise traditional methods of multi- objective optimisation. Evolutionary Algorithms (EA), on the other hand, are often seen as the most suitable method for tackling the problems exhibiting such characteristics. Our present study proves this to be true and shows that EA are capable of finding solutions undetectable by other optimisation techniques. 1 Introduction Cancer chemotherapy is a highly complex process which controls tumour development by the administration of a cocktail of chemicals in a series of doses over a course of treatment. There is a wide variety of anti-cancer drugs available to oncologists. Due to their high toxicity, these drugs give rise to a variety of side- effects, ranging from cosmetically undesirable through debilitating through to the effects that are themselves life threatening. The oncologist therefore is faced with a complex task of designing a therapy which achieves certain treatment goals whilst limiting the toxic side-effects of the treatment to an acceptable level. In the treatment of most common cancers multi-drug combinations are usually used. Traditionally, combination treatments are developed through empirical trials of different combinations, dosing, schedules and sequencing. However, since around 35 drugs are in common clinical use nowadays [17], it is evident that an almost infinite number of treatment schedules are conceivable and that the need for the optimisation of chemotherapeutic treatment is indisputable. The number of combinatorial possibilities for multi-drug schedules, coupled to the conflicting nature [13] and non- linearity of the constraints imposed on cancer treatments, make it difficult to solve the