John Klein

Thomas Duboudin, Xavier Fontaine, Etienne Andrier et al.

Computational pathology has made significant progress in recent years, fueling advances in both fundamental disease understanding and clinically ready tools. This evolution is driven by the availability of large amounts of digitized slides and specialized deep learning methods and models. Multiple self-supervised foundation feature extractors have been developed, enabling downstream predictive applications from cell segmentation to tumor sub-typing and survival analysis. In contrast, generative foundation models designed specifically for histopathology remain scarce. Such models could address tasks that are beyond the capabilities of feature extractors, such as virtual staining. In this paper, we introduce CytoSyn, a state-of-the-art foundation latent diffusion model that enables the guided generation of highly realistic and diverse histopathology H&E-stained images, as shown in an extensive benchmark. We explored methodological improvements, training set scaling, sampling strategies and slide-level overfitting, culminating in the improved CytoSyn-v2, and compared our work to PixCell, a state-of-the-art model, in an in-depth manner. This comparison highlighted the strong sensitivity of both diffusion models and performance metrics to preprocessing-specific details such as JPEG compression. Our model has been trained on a dataset obtained from more than 10,000 TCGA diagnostic whole-slide images of 32 different cancer types. Despite being trained only on oncology slides, it maintains state-of-the-art performance generating inflammatory bowel disease images. To support the research community, we publicly release CytoSyn's weights, its training and validation datasets, and a sample of synthetic images in this repository: https://huggingface.co/Owkin-Bioptimus/CytoSyn.

Rui Min, Christelle Garnier, François Septier et al.

The particle filter (PF) is a powerful inference tool widely used to estimate the filtering distribution in non-linear and/or non-Gaussian problems. To overcome the curse of dimensionality of PF, the block PF (BPF) inserts a blocking step to partition the state space into several subspaces or blocks of smaller dimension so that the correction and resampling steps can be performed independently on each subspace. Using blocks of small size reduces the variance of the filtering distribution estimate, but in turn the correlation between blocks is broken and a bias is introduced. When the dependence relationships between state variables are unknown, it is not obvious to decide how to split the state space into blocks and a significant error overhead may arise from a poor choice of partitioning. In this paper, we formulate the partitioning problem in the BPF as a clustering problem and we propose a state space partitioning method based on spectral clustering (SC). We design a generalized BPF algorithm that contains two new steps: (i) estimation of the state vector correlation matrix from predicted particles, (ii) SC using this estimate as the similarity matrix to determine an appropriate partition. In addition, a constraint is imposed on the maximal cluster size to prevent SC from providing too large blocks. We show that the proposed method can bring together in the same blocks the most correlated state variables while successfully escaping the curse of dimensionality.

Gilles Wainrib, Barbara Bodinier, Haitem Dakhli et al.

Recent work has questioned whether large language models (LLMs) can perform genuine in-context learning (ICL) for scientific experimental design, with prior studies suggesting that LLM-based agents exhibit no sensitivity to experimental feedback. We shed new light on this question by carrying out 800 independently replicated experiments on iterative perturbation discovery in Cell Painting high-content screening. We compare an LLM agent that iteratively updates its hypotheses using experimental feedback to a zero-shot baseline that relies solely on pretraining knowledge retrieval. Access to feedback yields a $+53.4\%$ increase in discoveries per feature on average ($p = 0.003$). To test whether this improvement arises from genuine feedback-driven learning rather than prompt-induced recall of pretraining knowledge, we introduce a random feedback control in which hit/miss labels are permuted. Under this control, the performance gain disappears, indicating that the observed improvement depends on the structure of the feedback signal ($+13.0$ hits, $p = 0.003$). We further examine how model capability affects feedback utilization. Upgrading from Claude Sonnet 4.5 to 4.6 reduces gene hallucination rates from ${\sim}33\%$--$45\%$ to ${\sim}3$--$9\%$, converting a non-significant ICL effect ($+0.8$, $p = 0.32$) into a large and highly significant improvement ($+11.0$, $p=0.003$) for the best ICL strategy. These results suggest that effective in-context learning from experimental feedback emerges only once models reach a sufficient capability threshold.

Arnaud Deleruyelle, Cristian Versari, John Klein

Background: Accurate segmentation of microscopic structures such as bio-artificial capsules in microscopy imaging is a prerequisite to the computer-aided understanding of important biomechanical phenomenons. State-of-the-art segmentation performances are achieved by deep neural networks and related data-driven approaches. Training these networks from only a few annotated examples is challenging while producing manually annotated images that provide supervision is tedious. Method: Recently, self-supervision, i.e. designing a neural pipeline providing synthetic or indirect supervision, has proved to significantly increase generalization performances of models trained on few shots. The objective of this paper is to introduce one such neural pipeline in the context of micro-capsule image segmentation. Our method leverages the rather simple content of these images so that a trainee network can be mentored by a referee network which has been previously trained on synthetically generated pairs of corrupted/correct region masks. Results: Challenging experimental setups are investigated. They involve from only 3 to 10 annotated images along with moderately large amounts of unannotated images. In a bio-artificial capsule dataset, our approach consistently and drastically improves accuracy. We also show that the learnt referee network is transferable to another Glioblastoma cell dataset and that it can be efficiently coupled with data augmentation strategies. Conclusions: Experimental results show that very significant accuracy increments are obtained by the proposed pipeline, leading to the conclusion that the self-supervision mechanism introduced in this paper has the potential to replace human annotations.

Nathan Bigaud, Vincent Cabeli, Meltem Gürel et al.

While large language models (LLMs) are rapidly advancing scientific research, they continue to struggle with core biological reasoning tasks essential for translational and biomedical discovery. To address this limitation, we created and curated eight comprehensive benchmark datasets comprising over 300,000 verifiable question-and-answer pairs, each targeting critical challenges in drug discovery including target druggability, modality suitability, and drug perturbation effects. Using this resource, we developed the OwkinZero models by post-training open-source LLMs through a Reinforcement Learning from Verifiable Rewards strategy. Our results demonstrate that specialized 8-32B OwkinZero models substantially outperform larger, state-of-the-art commercial LLMs on these biological benchmarks. Remarkably, we uncover evidence of a key aspect of generalization: specialist models trained on a single task consistently outperform their base models on previously unseen tasks. This generalization effect is further amplified in our comprehensive OwkinZero models, which were trained on a mixture of datasets and achieve even broader cross-task improvements. This study represents a significant step toward addressing the biological reasoning blind spot in current LLMs, demonstrating that targeted reinforcement learning on carefully curated data can unlock generalizable performance in specialized models, thereby accelerating AI-driven biological discovery.

John Klein, Mahmoud Albardan, Benjamin Guedj et al.

We examine a network of learners which address the same classification task but must learn from different data sets. The learners cannot share data but instead share their models. Models are shared only one time so as to preserve the network load. We introduce DELCO (standing for Decentralized Ensemble Learning with COpulas), a new approach allowing to aggregate the predictions of the classifiers trained by each learner. The proposed method aggregates the base classifiers using a probabilistic model relying on Gaussian copulas. Experiments on logistic regressor ensembles demonstrate competing accuracy and increased robustness in case of dependent classifiers. A companion python implementation can be downloaded at https://github.com/john-klein/DELCO

Arthur Pignet, Chiara Regniez, John Klein

Despite the increasing demand for safer machine learning practices, the use of Uncertainty Quantification (UQ) methods in production remains limited. This limitation is exacerbated by the challenge of validating UQ methods in absence of UQ ground truth. In classification tasks, when only a usual set of test data is at hand, several authors suggested different metrics that can be computed from such test points while assessing the quality of quantified uncertainties. This paper investigates such metrics and proves that they are theoretically well-behaved and actually tied to some uncertainty ground truth which is easily interpretable in terms of model prediction trustworthiness ranking. Equipped with those new results, and given the applicability of those metrics in the usual supervised paradigm, we argue that our contributions will help promoting a broader use of UQ in deep learning.

Barbara Bodinier, Gaetan Dissez, Lucile Ter-Minassian et al.

High-throughput preclinical perturbation screens, where the effects of genetic, chemical, or environmental perturbations are systematically tested on disease models, hold significant promise for machine learning-enhanced drug discovery due to their scale and causal nature. Predictive models trained on such datasets can be used to (i) infer perturbation response for previously untested disease models, and (ii) characterise the biological context that affects perturbation response. Existing predictive models suffer from limited reproducibility, generalisability and interpretability. To address these issues, we introduce a framework of Layered Ensemble of Autoencoders and Predictors (LEAP), a general and flexible ensemble strategy to aggregate predictions from multiple regressors trained using diverse gene expression representation models. LEAP consistently improves prediction performances in unscreened cell lines across modelling strategies. In particular, LEAP applied to perturbation-specific LASSO regressors (PS-LASSO) provides a favorable balance between near state-of-the-art performance and low computation time. We also propose an interpretability approach combining model distillation and stability selection to identify important biological pathways for perturbation response prediction in LEAP. Our models have the potential to accelerate the drug discovery pipeline by guiding the prioritisation of preclinical experiments and providing insights into the biological mechanisms involved in perturbation response. The code and datasets used in this work are publicly available.

Arthur Pignet, John Klein, Genevieve Robin et al.

Deploying digital pathology models across medical centers is challenging due to distribution shifts. Recent advances in domain generalization improve model transferability in terms of aggregated performance measured by the Area Under Curve (AUC). However, clinical regulations often require to control the transferability of other metrics, such as prescribed sensitivity levels. We introduce a novel approach to control the sensitivity of whole slide image (WSI) classification models, based on optimal transport and Multiple Instance Learning (MIL). Validated across multiple cohorts and tasks, our method enables robust sensitivity control with only a handful of calibration samples, providing a practical solution for reliable deployment of computational pathology systems.

Arnaud Deleruyelle, John Klein, Cristian Versari

In practice, data augmentation is assigned a predefined budget in terms of newly created samples per epoch. When using several types of data augmentation, the budget is usually uniformly distributed over the set of augmentations but one can wonder if this budget should not be allocated to each type in a more efficient way. This paper leverages online learning to allocate on the fly this budget as part of neural network training. This meta-algorithm can be run at almost no extra cost as it exploits gradient based signals to determine which type of data augmentation should be preferred. Experiments suggest that this strategy can save computation time and thus goes in the way of greener machine learning practices.

Mahmoud Albardan, John Klein, Olivier Colot

We investigate a problem in which each member of a group of learners is trained separately to solve the same classification task. Each learner has access to a training dataset (possibly with overlap across learners) but each trained classifier can be evaluated on a validation dataset. We propose a new approach to aggregate the learner predictions in the possibility theory framework. For each classifier prediction, we build a possibility distribution assessing how likely the classifier prediction is correct using frequentist probabilities estimated on the validation set. The possibility distributions are aggregated using an adaptive t-norm that can accommodate dependency and poor accuracy of the classifier predictions. We prove that the proposed approach possesses a number of desirable classifier combination robustness properties.

John Klein, Hans van Vliet

Context: The systems comprising a system of systems (SoS) are inde- pendently acquired, operated, and managed. Frequently, the architecture documentation of these existing systems addresses only a stand-alone perspective, and must be augmented to address concerns that arise in the integrated SoS. Objective: We evaluated an architecture documentation viewpoint to address the concerns of a SoS architect about a constituent system, to support SoS design and analysis involving that constituent system. Method: We performed an expert review of documentation produced by applying the viewpoint to a system, using the active review method. Results: The expert panel was able to used a view constructed using the baseline version of the viewpoint to answer questions related to all SoS architect concerns about a constituent system, except for questions concerning the interaction of the constituent system with the platform and network infrastructure. Conclusions: We found that the expert panel was unable to answer certain questions because the baseline version of the viewpoint had a gap in coverage related to relationship of software units of execution (e.g., processes or services) to computers and networks. The viewpoint was revised to add a Deployment Model to address these concerns, and is included in an appendix.

George Mathew, Tim Menzies, Neil A. Ernst et al.

When Requirements Engineering(RE) models are unreasonably complex, they cannot support efficient decision making. SHORT is a tool to simplify that reasoning by exploiting the "key" decisions within RE models. These "keys" have the property that once values are assigned to them, it is very fast to reason over the remaining decisions. Using these "keys", reasoning about RE models can be greatly SHORTened by focusing stakeholder discussion on just these key decisions. This paper evaluates the SHORT tool on eight complex RE models. We find that the number of keys are typically only 12% of all decisions. Since they are so few in number, keys can be used to reason faster about models. For example, using keys, we can optimize over those models (to achieve the most goals at least cost) two to three orders of magnitude faster than standard methods. Better yet, finding those keys is not difficult: SHORT runs in low order polynomial time and terminates in a few minutes for the largest models.