The Integrated Science Lab invites you to join the conversation at a Lunch Pitch. Disruptive Ventures want to talk to researchers about commercializing their work. Kemal Avican wants to connect with research laboratories that conduct in vivo infection studies or metatranscriptomics. Christer Malm wants help to create models for training and validating label-free -omics data to accurately classify and sort biomarkers.
To encourage cross pollination of ideas between researchers from different disciplines, IceLab hosts interdisciplinary research lunches with the vision of allowing ideas to meet and mate. During the Lunch Pitch Season, the creative lunches take place at KBC (Glasburen) on a Wednesday.
Register to come to the pitch and reserve your lunch by Monday, 1 June at 10am.
IceLab Lunch Pitch registration will open two weeks before the event.
Note! The default lunch option is a vegetarian falafel sandwich. You can choose an alternative lunch in a separate form that will be emailed to you once you have registered.
Pre-seed investment and operational support for researchers commercializing their work
Abstract: Disruptive Ventures is a pre-seed investor focused on Swedish deep tech and B2B tech. We typically invest 0-1 MSEK at the earliest stage, sometimes before a company has been incorporated. We back it up with a team of experienced operators and company builders who work hands-on with our portfolio, primarily on the first commercial steps that turn a discovery into a company with real customers. We also support across other areas needed to build a successful company, such as recruitment, branding, financing, IP strategy, and soft funding applications. We are not researchers ourselves, our role is to advise and support on everything that surrounds the science once a company starts to take shape, so that the best researchers can keep doing what they do best. We have backed two UmU-rooted companies so far: InvivoRNA and LunaLEC.
What we are interested in getting out of this pitch: We want to meet researchers at IceLab and across UmU who are considering commercializing their work, whether that is in the near term or several years away. Early and informal conversations are what we are looking for, no formal pitch or finished idea required. We are also keen to connect with researchers who know colleagues thinking along these lines.
Enriching pathogens RNA in infected samples
Topic: Universal host RNA depletion to enable high-resolution, cost-effective sequencing of pathogens directly from infected tissues.
The Question: How can we economically and accurately sequence the gene expression of low-abundance pathogens (bacteria and viruses) when they are drowned out by an overwhelming background of host RNA?
Abstract: When studying infections in vivo, pathogen RNA is severely masked by host RNA, which often comprises >99% of the sample. This "needle-in-a-haystack" problem makes standard RNA sequencing economically unfeasible. To solve this, we developed dephoRNA, a novel, universal host RNA depletion technology. Using a unique "self-probing" mechanism, dephoRNA generates a self-adapting cDNA probe pool directly from the infected sample to guide RNase H. This eliminates >95% of the host RNA across the full transcriptomic spectrum without the need for pre-designed, species-specific commercial probes. Validated in both human and animal models, this method enriches bacterial transcripts over 14-fold, perfectly preserves physiological gene expression profiles, and reduces necessary sequencing depth (and cost) by up to 92%. We are currently transitioning this validated invention into a scalable, Research Use Only pre-library preparation kit.
What I am interested in getting out of this pitch: We are looking to connect with academic, clinical, and veterinary research laboratories that conduct in vivo infection studies or metatranscriptomics.
The "Curse of Dimensionality" in -omic Biomarker Discovery
'-omics' research generates massive, high-dimensional datasets, but rarely translate into clinically viable diagnostics.
What is the issue?
We are faced with an extreme "p >> n" problem (where features far outnumber samples). An analysis of a single sample can yield >8 000 proteins and >50 000 peptides, while training cohorts typically consist of <100 subjects/samples. This creates a massive, noisy data space. When machine learning algorithms are applied, they inevitably suffer from severe over-fitting. The models memorize random biological variations, sample-handling artifacts, and random protein degradation instead of genuine disease signals.
As a result, predictive models frequently achieve AUC of >0.9. However, upon testing novel, independent validation datasets, the models collapse, with sensitivity dropping below 30%.
We are generating highly complex data that fails to generalize to unseen clinical samples.
Clinical vs. Legal Specificity
The thresholds required for clinical and diagnostic testing (50%) are substantially lower compared to legal threshold for anti-doping conviction (>99%).
The challenge
To apply strict validation techniques and feature selection to ensure the models do not over-fit.
What do I seek?
Data science and AI expertise to handle data processing and predictive modeling.
Models for training and validating label-free -omics data to accurately classify and sort biomarkers.
KBC Glasburen, near the KBC café. Find your way to the venue (mazemap link)
IceLab Lunch Pitches are made possible through funding from KBC for the venue and from Stress Response Modeling at IceLab for their coordination and lunches.
