A multi-class predictor based on a probabilistic model: application to gene expression profiling-based diagnosis of thyroid tumors

Background Although microscopic diagnosis has been playing the decisive role in cancer diagnostics, there have been cases in which it does not satisfy the clinical need. Differential diagnosis of malignant and benign thyroid tissues is one such case, and supplementary diagnosis such as that by gene expression profile is expected. Results With four thyroid tissue types, i.e., papillary carcinoma, follicular carcinoma, follicular adenoma, and normal thyroid, we performed gene expression profiling with adaptor-tagged competitive PCR, a high-throughput RT-PCR technique. For differential diagnosis, we applied a novel multi-class predictor, introducing probabilistic outputs. Multi-class predictors were constructed using various combinations of binary classifiers. The learning set included 119 samples, and the predictors were evaluated by strict leave-one-out cross validation. Trials included classical combinations, i.e., one-to-one, one-to-the-rest, but the predictor using more combination exhibited the better prediction accuracy. This characteristic was consistent with other gene expression data sets. The performance of the selected predictor was then tested with an independent set consisting of 49 samples. The resulting test prediction accuracy was 85.7%. Conclusion Molecular diagnosis of thyroid tissues is feasible by gene expression profiling, and the current level is promising towards the automatic diagnostic tool to complement the present medical procedures. A multi-class predictor with an exhaustive combination of binary classifiers could achieve a higher prediction accuracy than those with classical combinations and other predictors such as multi-class SVM. The probabilistic outputs of the predictor offer more detailed information for each sample, which enables visualization of each sample in low-dimensional classification spaces. These new concepts should help to improve the multi-class classification including that of cancer tissues.

SEVERAL carcinogenic viruses, known to be responsible for spontaneous tumours in a given species, are able to induce a malignant transformation either in vitro or in vivo in the cells of an animal of another species. Yet their discovery has always been made by their inoculation into animals of the same species in which they had induced a so-called spontaneous tumour. This applies particularly to the Gross virus, the only virus whose role in spontaneous leukaemogenesis in mice is certain (Gross, 1961).
It would seem that the greatest chance of revealing a leukaemogenic virus, in a material which only contains a small amount, would be by injecting animals of the same species as those in which it had appeared, rather than by injecting into other species. It is reasonable to think that there would be a greater chance of revealing a possible leukaemogenic virus in man if the biological material being studied were to be injected into human beings rather than into animals. These experiments in a direct form are not ethically acceptable, but an indirect approach could be used by injecting human material into animals with functioning human haemopoietic tissue-animal human chimeras.
Furthermore, it must be remembered that the detection of viruses responsible for spontaneous tumours is often facilitated by using immuno-incompetent hosts, which are usually new-born animals. In haematopoietic chimeras there also exists an immune incompetence (Mathe, Amiel and Daguet, 1961), which is another reason for using this system. Before attempting to form haemopoietic chimeras between man and monkeys, we studied the possibility of detecting a known leukaemogenic virus in mice by inoculation of rat-mice haemopoietic chimeras. This was preceded by experiments in which the leukaemogenic viruses of the Friend (1957) or Rauscher (1962) type were demonstrated by their injection into animals with an allogeneic chimera of the haemopoietic system (Mathe and Amiel, 1966).

MATERIALS AND METHODS
Four hundred and twenty rats, male and female, of the Wistar CF strain, aged 8 days, were irradiated with a dose of 800 rads total body irradiation (260 kv 0.5 mm./Cu, CDA 1-75 mm./Cu, distance 70 cm.). On the following day they were injected intravenously with 1-5 x 108 bone marrow cells from (DBA/2 x C57B1/6) F1 hybrid mice (males or females) and 6 days after this transfusion they were injected intraperitoneally with 0-2 ml. of a solution of Friend virus (the supernatant fronm a 35 per cent w/v solution of spleen, centrifuged at 2500 r.p.m. for 20 minutes).
When the rats died spontaneously or when they were killed because their death appeared to be imminent, the chromosomes of bone marrow and the spleen were examined and a full macroscopic and mlicroscopic examination made of the animals. The cytotoxicity of rat antisera against mouse cells and of mouse anti-sera against rat cells was studied on the bone marrow and spleen. The percentage of cells killed by each of these anti-sera in the same cell suspension allowed the percentage of each cell type in the mixture to be calculated. The technique used was that described by Bennett, Old and Boyse (1964). In each test the cytotoxic action of each anti-serum was checked on pure populations of lymphoid cells from rats and mice of the same strains as used in the chimeras. The antisera were taken from animals of the same strains as those used for the heterospecific graft experiments.
When splenomegaly was observed 106 spleen cells were inoculated into CF Wistar rats and (DBA/2 x BALB/c) F1 hybrid mice of the opposite sex to the animals that had served as the donor of bone marrow to the rat on which the genetic identity of the spleen cells had been studied.

RESULTS
Three hundred and twenty one rats died of aplasia, 57 died of the runt syndrome ( Fig. 1) between the 20th and 50th day and in 42 animals leukaemia was observed between the 14th and 81st day (Fig. 2).
The establishment of a haemopoietic cell heterograft was proved easily as 100 per cent of the bone marrow cells in mitosis had mouse chromosome (Fig. 3), and by the study of the antigenicity of the marrow cells using the cytotoxic anti-sera. The runt syndrome provided further evidence of the active function of the graft.
Several pieces of evidence pointed to the fact that the leukaemic cells were of murine origin. The leukaemnia was a hepatosplenomegalic leukaemia (Fig. 2) identical to that induced by the Friend virus in mice. It was an erythroblastic and monoblastic leukaemia and not a lymphocytic leukaemia. In the leukaemic animals, most of the cells in the spleen were involved in the leukaeniic process and these were shown to be of murine type by chromosomal studies (100 per cent of the chromosomes were shown to be of the mouse type) and by the cytotoxicity test (Table I). Further evidence of the murine origin of the leukaemic cells was provided by the grafting experiments in which leukaemic cells from the spleens of the rat-mice chimeras, which received marrow grafts from female (DBA/2 x BALB/c) F1 hybrid mice, were injected into adult intact CF Wistar rats and adult intact male (DBA/2 x BALB/c) F1 hybrid mice. None of the 67 rats but all of the 18 mice developed leukaemia, and chromosome analysis of EXPLANATION OF PLATES. FiG. 1.-Wistar CF rats of the same age. The two smaller animals are suffering from runt disease following irradiation and grafting of haemopoietic cells from (DBA/2 x BALB/c)  the spleen cells from the leukaemic mice showed that the majority were of female karyotypes (Table II). This experiment showed that the leukaemia in the mice was largely due to the graft and not induced by the virus carried by the leukaemic cells, as the recipients (all males) developed a leukaemia in which the majority of the cells did not have a Y chromosome which can easily be identified (Fig. 4). This demonstrates that it is possible for cells, induced to become leukaemic by the Friend virus, to establish themselves as a graft, as well as inducing leukaemia in the cells of the grafted host due to the virus present in the leukaemie cells, In our experiments the cells of the donor and the host could be distinguished by their sex chromosomes.
The system we have proposed would appear to be a very sensitive method to detect a leukaemogenic virus in a species, without having to use animals of that species and cause them to runi the risk of contracting the leukaemia. We now consider that this system merits use in the search for leukaemogenic viruses in man, using a heterochimera of the human haemopoietic cells in a monkey. SUMMARY A system is proposed which uses haemopoietic chimeras to detect a leukaemogenic virus. The present work has demonstrated that Friend virus can be demonstrated in a rat-mouse heterospecific chimera. It is suggested that chimeras of 147 human haemopoietic cells in the monkey might be used to detect leukaemogenic viruses in man.
This work was done with the aid of INSERM, contract no. CR.66-235.